channel_id stringclasses 1
value | video_id stringclasses 1
value | channel_name stringclasses 1
value | kind stringclasses 1
value | title stringlengths 10 65 | description stringlengths 118 5k | num_views stringlengths 1 3 | num_likes stringclasses 11
values | num_comments stringclasses 4
values | transcript stringlengths 733 71.6k |
|---|---|---|---|---|---|---|---|---|---|
UCZbtcRQHGpU_tt2yM4C7k0A | NRrH9-kRMwc | Changelog | youtube#video | Broccoli AI at its best 🥦 | We discussed “🥦 Broccoli AI” a couple weeks ago, which is the kind of AI that is actually good/healthy for a real world business. Bengsoon Chuah, a data scientist working in the energy sector, joins us to discuss developing and deploying NLP pipelines in that environment. We talk about good/healthy ways of introducing AI in a company that uses on-prem infrastructure, has few data science professionals, and operates in high risk environments.
Leave us a comment (https://changelog.com/practicalai/280/discuss)
Changelog++ (https://changelog.com/++) members save 5 minutes on this episode because they made the ads disappear. Join today!
Sponsors:
• Intel Innovation 2024 (https://intel.com/innovation?regcode=CMCCHL&utm_campaign=Changelog) – Early bird registration is now open for Intel Innovation 2024 in San Jose, CA! Learn more (https://intel.com/innovation?regcode=CMCCHL&utm_campaign=Changelog) OR register (https://reg.oneventseries.intel.com/flow/intel/innv2024/InnovationReg?regcode=CMCCHL&utm_campaign=Changelog)
• Motific (https://www.motific.ai/) – Accelerate your GenAI adoption journey. Rapidly deliver trustworthy GenAI assistants. Learn more at motific.ai (https://www.motific.ai/)
Featuring:
• Bengsoon Chuah – Twitter (https://twitter.com/bengsoon) , GitHub (https://github.com/bengsoon) , LinkedIn (https://www.linkedin.com/in/bengsoon)
• Daniel Whitenack – Twitter (https://twitter.com/dwhitena) , GitHub (https://github.com/dwhitena) , Website (https://www.datadan.io/)
Show Notes:
• MLFlow (https://mlflow.org/)
• Prefect (https://www.prefect.io)
• DuckDB (https://duckdb.org/)
• Agrilla (https://argilla.io/)
Something missing or broken? PRs welcome! (https://github.com/thechangelog/show-notes/blob/master/practicalai/practical-ai-280.md) | 269 | 3 | 0 | [Music], welcome to practical AI if you work in, artificial intelligence aspire to or are, curious how AI related Tech is changing, the world this is the show for you thank, you to our partners at fly.io the home, of, changelog.md 30 plus regions on six, continents so you can launch your app, near your users learn more at, [Music], fly.io what's up friends Intel, Innovation 2024 is right around the, corner accelerate the future, registration is now open and it takes, place September 24th and 25th in San, Jose California this event is all about, you the developer the community, and the critical role you play in, tackling the toughest challenges across, the industry ignite your passion for AI, and Beyond grow your skills to maximize, your impact and network with your peers, as they unleash the next wave of, advancements in technology here's what, you can expect understand the emerging, Innovation and Trends in Dev tools, languages Frameworks and Technologies in, Ai and Beyond to empower you and the, solutions You're Building get in-depth, technical experence experience doing, handson workshops Labs meetups and, hackathons to collaborate and solve, problems in real time you can explore, featured partner and Intel solutions, they have Partners there startups there, customers there and Intel is showcasing, the latest in products services and, solutions across Keynotes Tex sessions, and the show floor to help you meet your, development needs collaborate with, experts learn and have fun engage in, interactive sessions to connect get, certified gain unique ideas and, perspectives build longl lasting, networks and of course have fun and get, inspired hear from leading industry, experts technologists startup, entrepreneurs and fellow developers, along with Intel leadership CEO Pat, gillinger and CTO Greg lavender as they, take you through the latest advancements, in technology don't miss this chance to, be at the Forefront of innovation take, advantage of early bird pricing right, now until August 2nd register using the, link in our show notes or to learn more, go to, intel.com Innovation once more that's, intel.com inovation or go to the show, notes and click that, [Music], link welcome to another episode of the, Practical AI podcast this is Daniel Whit, neck I am founder and CEO at prediction, guard and this is a this is a pretty, special and uh and fun episode for me, because I get to kick back with an old, friend of mine we uh we went to the same, University for those that haven't heard, of it Colorado School of minds and in, Golden Colorado of course a shout out to, uh all the or diggers out there that are, that are listening but yeah we we have, uh we have with us today bingen Chua, who's a data scientist and now um and, working in the in the energy sector um I, was really fascinated to talk over the, years with with bingson about uh about, all the things he's doing and in, particular uh his kind of approach and, learnings around Active Learning and NLP, models and yeah I wanted to invite him, on the show to talk through some of that, and learn a little bit from him so, welcome to the show how you doing hi, Daniel thanks for having me yeah it's, been a while since the days in Colorado, School of Minds in, in and now uh you're working a data, scientist in the energy sector and also, working in in Asia which is super cool, I'm wondering if you could give us a, little bit of a sense of some of the, unique things about doing data science, and machine learning type of things in, the context of like the energy sector in, the context of like an actual Enterprise, real world kind of situation because we, talk a lot about recently we've been, talking a lot about all of these geni, models and apis and such and that is, super cool but also there's a lot of on, the ground work going on in data science, that maybe looks quite a bit different, than that so yeah thanks um so I mean I, I work in in the energy sector and uh, it's pretty much a tra traditional type, of sector a lot of the companies um as, you go around at least in Asia or at, least over here where I'm at we do not, actually even have things like cloud, services or you know subscription and, stuff like that due to different reasons, and stuff but at the same time there's, an appetite for machine learning AI data, and all of those things um you see, people talk about geni as well but I, think the ones that I've noticed at, least for me personally that has really, brought a lot of values are kind of what, you guys were talking about in the, previous episode of a brockoli AI, broccoli AI I love, it yeah not so sexy but uh but still, really important really brings value, particularly I guess what we're going to, talk about is is active learning in the, context NLP natural language processing, and and so I think that's a that's a, pretty exciting place to be in um to to, do you know I mean it kind of translates, into genan to a certain degree but at, the same time I think to at least the, context that I'm in as well in in the, sector that I'm in we do not have things, like Health Services is um ready for us, right and so you have to figure out ways, to kind of bring that about um in a an, on Prem server BM so how do you work, around with that and how do you actually, um bring in Cloud native modern, Technologies within a traditional kind, of a structure so yeah yeah and from at, least my impression even though there's, sort of not this whether it be for, security reasons or Legacy reasons or, just connectivity that you know there's, there's not the type of connection to, cloud services like you're talking about, that others might be working with but at, the same time at least my my impression, is that this sort of sector and maybe, others there's other related verticals, where they have been sort of data driven, to some degree for some time and I don't, know if you could speak to that like the, the types of data that people you know, have been processing or storing or are, available in in those contexts but yeah, I mean that's a good point because um a, lot of these traditional Industries at, least you know in the energy sector that, we see we have sensors that are, constantly flowing in data all the time, the data is there I mean we we are, collecting data right and then going a, little bit deeper then you find that a, lot of us have been collecting a lot of, unstructured data too and so at least, within um you know where in my, experience at least what what was, happening was when I came in I was, pretty much the only data scientist and, stuff and so I had to like make my um, existence justified in a sense and so I, knew that like I had to do something, about like bring value within this, organization and that like I'll be able, to kind of prove that like hey look data, science actually does work and does, bring value but the quickest I guess lwh, hanging fruit that I found within at, least in the context of where I'm at is, is the whole unstructured data so we, have been collecting thousands and, thousands not hundreds of thousands of, unstructured data but there has never, been a way to really Analyze That at, scale so people have been analyzing it, they've been able to do some sort of a, human analysis on it but there's never, been someone who's able to like say hey, what's been happening in the past 10, years that we've been collecting all, this data what's it telling you um, nobody has really been able to do that, so I thought okay you know maybe that, could be one thing that we could, actually bring in is that like it you, have all this data that's ready for you, that has all the inside that has all the, information that is locked up right, waiting to be on earth pretty much, waiting there for us to just extract it, or mine it um so I I I did a quick PC, for one of the the Departments company, and uh said hey guys like you've been, doing a powerbi tableau powerbi kind of, thing you've been able to do stuff with, your um structured data right and you're, being able to plot it on beautiful, graphs and stuff like they be able been, able to analyze in that sense what about, all the instructure data that you've, been collecting over the years and they, said there's no way we could do it right, I mean like we've got thousand how you, can't there's no possible way for you, put it in RBI and so I said like well, maybe we could explore ways that we, could actually get into machine learning, to actually help you to scale that, analysis um from ANL standpoint, thankfully they bought it and that's, what we took off from there and uh it, was pretty cool I mean it was a journey, for sure journey to learn yeah so like, when you say unstructured data uh give, people a sense of like the kinds of, files or um not not maybe the specifics, but you know I'm imagining a file store, with some type of files in it that, contain yeah something give people a, sense of of that yeah for sure I mean I, guess unstructured data typically we, think about it as like text right um it, could be text or it could be something, that's just out of structure and in like, Microsoft the docs, or yeah so we have been storing all of, those data within SharePoint Microsoft, SharePoint right and so what I've seen, is these instructed data is usually it, usually comes in a tablate form right, you have a table that is collecting all, of the structured data but alongside, with it there's always that comment or, that additional things that you have to, actually tell the story of what you're, actually collecting and um those are the, ones that I think it's always Loft up, and you would see it's it's very typical, in a any kind of industry where you have, a tablet data data that collects you, know sensor data and everything or like, reports of what's been happening and, those are like structured but then, there's also a column that would bring, in some sort of remarks you know, observation or actions taken whatever it, is comments comments yeah but um more, often than not we just kind of like play, through it and and kind of like not, really put too much attention to it at, least within the data set we were, looking at um we had found that there, were lot more insight in those data than, the structured data that they've been, collecting I'm talking about like safety, data so we've been collecting like, safety reports every single day couple, hundred sometimes or tens of them every, day over the years and so these data has, some sort of a Insight right and it, brings in an Insight of how is a safety, condition um you know operations and so, with that like they people usually see, at the categories or that that what the, reports are being reported for but then, like when you look at the categories you, realize that sometimes it doesn't really, jive with the stories that they're, actually trying to bring in in, instructure data and so that's where the, the part where we felt like it's going, to bring in additional insight to the, instructure data that we see yeah and so, like there's this uh we've talked a, little bit about like the data that you, was there the potential insights in that, around safety and maybe other insights, as you kind of came into this industry, and kind of we getting an understanding, of like let's say that you built the, coolest data science app that was out, there and and had a cool model in it to, do some analysis like what is the, reality of how that would have to run, like what is in production mean uh for, for in in your context yeah so many, things I mean just start from the data, we don't even have labeled data to start, with say you want a classification app, or model you need to have some sort of a, label right and uh we don't have that, and so you have to figure out ways to, bootstrap your labeling process to start, off and stuff and then all the way down, to like of course training the model is, pretty easy these days right and um but, then you have to think about things like, hey what does that look like to put it, on on the server most of our servers are, running Windows server and so I've had, experiences putting production some apps, on on Windows server and that was, painful and U so like we have to figure, out ways that like work with the it and, stuff and said hey can you deploy a a, Linux server for us instead and just, work it out from there and set it up, from there that being said like that, that's just over all picture of it and, then you get into details of like how do, you actually store your model you've got, to have some sort of a infrastructure to, kind of hold that which you know in our, case um felt like mlflow is pretty good, model registry experimentation tracking, and stuff to keep track of what type of, models that I'm using and stuff like, that and then so many things, honestly gosh and then like how do you, actually put it on an orchestration kind, of a Serv, you could use KRON jobs but then you, know it may not be so flexible then you, kind of need to work something out and, so you have to get some sort of, orchestrator to spin it up and and kind, of like make that a service for your, infrastructure as well I love this, discussion because it uh I think it fits, the theme of the show so well around, being practical and the fact that yeah, I'm sure that there's actually a good, number of listeners out there who are, really wanting to do, machine learning AI data science type of, things and they are sort of in a similar, situation in in their company because, actually I think probably more of the, majority of companies are in this sort, of situation than sort of infrastructure, wise extremely modern and just cranking, everything out on on kubernetes in the, in the cloud that sort of thing um so, yeah I I I love this um so you were, talking a little bit about the sort of, problem of so you have have all this, tabular data with extra comments and and, unstructured data and you know certain, things you want to do like extract, insights or classify maybe some of the, unstructured data but then also nothing, has been labeled over time it's just, unstructured data so talk a little bit, about that bootstrapping problem and and, how you've thought about that in terms, of I've got all this stuff I want to, create a model but I have no no starting, point when we were work going through, that whole labeling process or data, preparation process was pretty, interesting because we really didn't, have anything no labelers or anything, and um I didn't have a budget to get an, external labeler and for me know just, hire hire thousands of of people online, right I know I could I could just do, that maybe but then again even that like, I've thought of it and um our data, sensitive to start with but at the same, time it's so Nu on and it's it's so, Nuance to the context of our company and, so a lot of data that we that every, company has is just so new ones to their, own context you know and um at the same, time to one of the new ones is is is the, way that like um these texts are being, written and so a lot of quote switching, happening you know um which means quote, switching which means like um in Asia a, lot of times we speak in English but we, will kind of put in some you know native, languages that we know or that we grew, up with and so it's just kind of like, you go back and forth back and forth and, it's just kind kind of a common thing in, especially in Southeast Asia and um so, you can't just hire somebody online and, and just kind of label it for you, because you just can I don't even know, how to bring in that context of these, guys right and um but thankfully I mean, I would say that um the part that really, helped was I have to have a really good, sponsor for the project and these, sponsors um they were super on board, they were not technical um they were, smmes in their own, departments and they knew their stuff, but they know enough of machine learning, and data and AI that hey it's you know, kind of a model that predicts not, necessarily making 100% accuracy get-go, and so they understand those nuances in, a sense and so they kind of supported, that and understood the kind of uh, things that we have to go through you, know as a practitioner that we have to, go through they kind of understand that, part of it um so that was really helpful, for my part because having a good, sponsor means you get really good, support for the project but that also, means that um because the product that, I'm working the app that I'm working for, on is for their people you know their, subordinates and the people and people, were reporting under them and what, happens is they said to them and say hey, guys this is your app I want you to help, B soon out with building this app and so, having the users themselves on board, from the foundational bootstrap level, really helped us so because we didn't, have any labels um they had the guys to, actually be the ones who label for us, they were the labelers and so the the, users themselves were the labelers so, that was that honestly I was pretty, really blessed to to even just have that, that kind of worked out together um I, think that that kind of works so much in, my favor yeah yeah and in that labeling, process how did you develop your sort of, set of instructions for like how you, like explaining the problem to them or, helping them Define the problem and the, categories for example in a, classification model how was it for you, because I've I've also had the, experience personally probably and been, burned a couple times where I'm like oh, this problem makes sense to me I set up, the labeling thing I release a bunch of, labelers in there and the either the, instructions don't make sense or I've, biased this in some way or you know, likely because I like you had mentioned, I wasn't super close maybe to the to the, users in in that situation but um yeah, any learnings from that experience I, think um just a lot of iteration with, them um I had so many times like I would, travel to see them um they work in our, operations so I literally travel there, um to see them in person and I said I, would we would just go through hey these, are the labels that we want to label we, kind of get a general idea of what they, are but when you get into the weeds of, it like you when you get into the, details of it you're like you would, think in this situation that label, should be here should be number one but, then no somebody else said no it's two, so um the way I worked it out was there, will always be contention I notice no, matter how tightly nit your labelers or, your team are that will always be, contention and you just got to work, around with it at least to in my, experience um I just had to work around, with it and the way I worked around with, it was I just kind of had a voting, system you know and um I I set up an, account so the technical side of this is, I I could have just given them Excel, sheets and they could just label them on, an Excel shape but I find that um you, know they are doing put me a favor kind, of thing but I want them to have a, really good user experience instead of, just going through Excel sheets and, stuff and so I I used uh argila back in, the days and uh when they started and um, preh hugging face days yeah yeah exactly, and and I noticed that arila is amazing, in the sense that like it allows you to, set up different user, and you could I mean even other kind of, interfaces I've used labels to as well, but argila was able to I could use the, API and just kind of like set up each, user right and then for each user I'll, would just kind of like sample the same, data for them um so I had to actually go, through the first round the same number, of set of uh data for them to label say, 500 of them I think I remember and they, would all label within two weeks and at, the end of that two weeks I'll collect, them and I'll find which one which are, the most contentious ones and so the, ones that are the most contentious the, ones that have the least um percentage, of the majority I would pull it up and I, said hey guys what do you think about, this this is contentious for you guys, why is it contentious and you look it up, from there right because chances are, you're going to see the same kind of a, label again of the same kind of a data, again and um if you talk it out, hopefully when you see a similar thing, and you said hey I actually we already, talked about this we all agreed that, we're going to go with this and so, that's first round we had the same label, there the same data set for everyone, it's bit inefficient to a certain degree, but I think it's important to actually, get into that place so you understand, the contentions of each person and then, the second round I have everyone just, kind of like label at scale pretty much, and um yeah we collect that from there, pretty much interesting yeah so you did, um in this process you did a initial, sort of offline bootstrapping of of, labels right and did that so like, scale-wise like what sort of scale when, you're solving so here we're talking, about an NLP problem creating a, classification model on some labels of, of this unstructured data of course this, would vary by domain but sort of what, scale of labels did you shoot for when, you were doing that initial trying to, get to that place where you could start, up and and train your first model we did, um just about 1800 to 2,000 labels okay, or rows of data basically and then we, start training our first model that, probably means you're not training like, a 400 billion parameter model with with, 1,00 samples you don't even enough, infrastructure to be able train that, yeah no we don't so what does broccoli, AI model uh look like I mean this are, all texts right and so we were we were, going with like the simpler simplest you, can find on on hugging face at that time, I think at the time sentence, Transformers were really making it big, um you know for different reasons like, whether it's topic modeling or or, classification and at the same time too, I remember they came up with the set fit, model which is fine-tuning the sentence, Transformer which was honestly, revolutionary for me yeah amazing and um, I thought it was amazing that like it's, a you're able to do um something that, was meant for similarity but then you, could actually fine tune it for, classification and and with pretty good, performance um and it's supposed to be, something that is a few shot, classification model few shot kind of a, fine-tuning and so I thought 2000 should, be enough for me to start somewhere, right and in fact when I trained that um, I tried some other models but I think, sentence Transformers were the ones that, actually gave the best performance out, of all um it still wasn't that good you, know talking about like 60 something 70%, kind of thing in terms of F1 score but, when I talked to my sponsors about this, I said hey guys like you okay with like, me deploying this at like 60 70% and, they said No actually that's fine right, because um the objective for this was, number one to bring visibility of these, reports to the users because one of the, pain points that the they said was for, us to be able to know what people have, been reporting that at least in the past, 24 hours they had to get on SharePoint, and just different hoops and Loops to, try to find out you know filtering and, stuff but to be able to get that sent, out in the email with the classification, was already win and so I thought okay, let's do that but let's not stop that, right I mean we should actually create a, pipeline and that's where the active, learning comes in it it really helped, because uh I'm glad that I actually used, argila to start with the bootstrapping, of uh our data set and having the RG, which means our users already used to, the interface and they already have an, account and so I was able to kind of, hack around with the argila a Python, apepi and um basically I uh was able to, create a loop where pretty much what, this model does every day it will bring, in the new data that people have been, reporting for the last 24 hours and make, some prediction on on it at about 60 70%, F1 score accuracy whatever it is and, then send it out to the users and these, users will see it and at the end of that, email they would say hey um I don't, think this is that signal it should be, this signal I want to give a feedback, and at the end of it they're able to, click on a link that brings them to, their profile in argila that will allow, them to give it the feedback for the, particular day data set and so over time, now it's in production every day I would, get from time to time I'll get people, giving their feedback and we've gotten, like close to 4,000 um data sets now, label um from this Active Learning and, so we so we will train model, periodically not on what I could I could, have done it automated but I didn't, really want to like just I didn't feel, the need for it yet to put it on, animation um but then at the same time, like you know you you're just collecting, an Anin and we're just training it from, time to time, basically hey friends out shift Cisco's, incubation engine merges Innovation with, the art of possible a Launchpad for, transformative emerging Tech out shift, Blends startup agility with corporate, strength to develop nextg Technologies, from the groundup in AI Quantum, Technologies Cloud native and more their, newest AI Innovation Motif addresses a, critical challenge in the rapidly, advancing world of gen AI Bridging the, Gap between concept and deployment this, model and vendor agnostic solution, supports the entire gen AI Journey from, assessment and experimentation Motif, accelerates deployment from months to, days while safeguarding against gen AI, security trust compliance and cost risks, all while empowering business function, and it teams to rapidly configure and, user assistance powered by, organizational data Motif provides, Advanced customizable policy controls to, prevent unauthorized access to sensitive, data and helps ensure compliance, throughout the entire process with deep, visibility into operational and business, metrics motivic enables you to track all, why optimize costs and make informed, Decisions by offering a centralized view, moic deters Shadow AI usage and empowers, teams to innovate responsibly so move, beyond the traditional constraints of AI, implementation utilizing AI deployment, that is both responsible and is, revolutionary ensuring your projects are, not just quickly launched but built on a, foundation of trust and efficiency visit, motif a that is m o, tfic c., [Music], AI so Bings soon uh it's super, interesting to hear kind of how the, you were able to engage the the users of, the application through this like, reporting process essentially that they, were you know had some of the right, incentives in place to to respond and to, give you updated labels and you, mentioned also the model repository, saving models getting them out with, mlflow in the context of you know you, deploying your model on Prem you, updating the model you just mentioned, kind of retraining the model what does, that look like for you right now in, terms of that cycle of when you would, want to push out a new model after, gathering this data how you would judge, that to be worthwhile or or useful in, any sort of testing that that is, relevant to that cycle of getting in new, labels retraining you know evaluating, that sort of thing what what does that, look like for you and how do you kind of, put in the right or or how have you, thought about the right metrics to, understand when to update the model at, this point honestly we we keep it simple, we just kind of like periodically do it, at a Cadence you know a couple months or, two but I did think about like what does, it look like to actually measure the, drift of the data and stuff like that of, the model predictions that could be one, of the ways that we could do it too but, what I'm seeing is actually the model is, um doing its job fairly well well enough, to actually soft the the business, problem and so we don't see a need to, actually Implement more sophisticated U, monitoring unless we need to you know, that's that's where we're at with it, yeah and when you push your model sort, of like you update it um you had, mentioned the model repository how are, you shipping your model out to the, application because I think like you had, mentioned you know you only have so many, resources I think there's there's also a, lot of people out there in your, situation where I think it was Kristen, Lum on a previous episode she had talked, about kind of that data scientist out, there that is maybe one of very few or, the only data scientist in a potentially, a large organization and having to like, do all of these things they're not like, an mlops person they're not a model, trainer they're not a observability, person they're doing all of that right, so there are limitations to to you know, how much sophistication you can put in, place and I think that like some people, go way too far and they're like oh I'm G, to implement all of this stuff and it, actually makes their life as a, practitioner less uh Happy than than, than otherwise so yeah how have you, found that balance and like what does it, look like for you to do these Cycles in, terms of tooling and the things maybe, that You' like you say you mentioned you, thought at some point maybe it's relev, to implement some of this observability, stuff but maybe not yet or there's other, priorities so what does that look like, for you in terms of how you decide what, what level of sophistication is right, and how you push things out and I'm 100%, with that honestly because uh it is a, matter of priority my customers are, happy I'm happy and I'm not going to, like change what's you know what's good, oh I don't want to break what's been, working right so to speak but that being, said like I you know when it comes to, like all of that I think I have a, general idea of what would be the, minimum thing so now I'm working on some, other things as well like phenomenally, detection and stuff like that which, needs to be deployed so having gone, through that that kind of like set up a, like a Sero infrastructure for me to, know what kind of a infrastructure that, I'm going to be looking for for whatever, else that I'm working on and um at the, bare minimum I think model registry is, is super important and uh being able to, call the different versions that you've, been training and being able to track, that and being able to call it through, an API through a function you know MLF, flow has this great python connection, with it and so being able to do that is, it's just amazing I mean it keeps my, life sane right I don't have to like, figure out where I store my model pretty, much so I would be doing exactly the, same things with whatever I'm working on, next which I I've since moved on from, that project and I'm just kind of like, maintaining it that project account now, in maintenance and now I have to move on, to a different project to solve a, different part of the business you know, in that sense but that project kind of, set like I said set the foundation and, knowing what kind of things that needs, to be done so sorry I'm kind of going, ahead of myself so one is ml flow being, the most important thing for me um in in, terms of this sort of scenario the other, one is um orchestrator is also really, important having a really robust, orchestrator so for me I think preact, was perfect for me you know and um I was, able to do different things and stuff, yeah the types of things that you're, orchestrating or what types of things, you could do it real time with prefect, at the same time you could also be, listening and stuff but at the same time, you could also just running on schedule, calling different functions um sub, functions and things like that um so, that was really cool to be able to have, that that's pretty much what we do right, now we're not really going into, real-time monitoring yet until do that, then we'll I have to figure out, something else more sophisticated yeah, yeah yeah and and are you just uh, shipping your models sort of as part of, a Docker container or or something like, that pretty much yeah we do use uh, Docker containers and uh just so that we, can keep it contained in that sense yeah, that's awesome I think you had mentioned, in one of our conversations something, about uh duct, DB where does that fit into some of this, so the data that raw data that we get, from is from SharePoint but if you have, uh anyone who has any experience with, SharePoint in terms of like wrangling, and data stuff it's so painful yeah so I, I thought that would be good to actually, have some sort of middle layer mini lake, house or data Lake kind of thing and um, I don't want to bother my it guys too, much so I thought Doug DB is a great, thing for it right I don't need a VM for, it and um you can have an embedded SQL, service that you can use so that's being, pulled every day pulling that data into, du DB and dub will be the one that, actually cleans up the data preparing, the data to send it to the model and um, that becomes like a pipeline for me to, be able to um work around the whole, complexity of uh SharePoint really, honestly yeah I have personally found a, lot of use for for duck DB even in the, past uh yeah even in the past year on, the even on the more geni stuff where, you're doing sort of like uh text to SQL, or like queries and that sort of thing, and every company we're working with has, different crazy sets of data or, different configurations of this or that, and that layer of having a kind of, unified analytics layer but also not, that's sort of uh you know easy to pull, in to python easy to spin up easy to, like test with locally and then like, deploy with, yeah that's been really useful I, remember you talked about Lance DB you, know for rack and things like that and, yeah it's the same thing I love like, embedded database I think it just works, well you know and like it's kind of, scalable eventually you know and I think, I I really like that I think there was, one uh blog post I've always referred, back to because I I also went through, the you know you and I were at mines at, the same time and then like there was, like data science and then there was, like the big data period where everybody, was in Hadoop and Spark and and all this, stuff which I know uh a good number of, people still use spark but I there's a, blog post by the mother Ducker uh oh, yeah company yeah but I think the title, is Big Data is dead or something it, basically goes through some of the, discussion around like hey we all, thought we had big data but like the, actual query problem like the types of, queries that we need to run these aren't, like big data problems what's needed is, different so yeah for those shout out to, whoever wrote that blog post because it, was really really good if if you ever, want to come on the show and talk about, it that that would be awesome yeah well, as you kind of look back on this process, and some of the things that that you've, learned like what are you what are you, looking forward to in terms of like the, future of the process of your own work, or or of the things you're learning or, maybe like as you go into this next, phase it sounds like you're working on, some new things you'll want to reuse, some of the tooling and kind of process, that you have used but you know what's, different or what are you excited about, kind of for this next phase in light of, what you've kind of learned over the the, past years generally speaking I think, mlops is just so nuanced you know in, different context everyone has this a, say of what should be done and I think, um if I learned something from this was, uh nobody really knows everything so you, kind of have to figure out from there, and you you kind of take a risk on, certain things that you decide in terms, of your system design and stuff what I'm, excited for is actually to be able to, take this and and see what it looks like, to for other things right and um in, other applications like whether it's, anomaly detection or whatever it is in a, broader sense I think I'm excited to to, see things like embedded data base you, know getting more and more mainstream, especially in the context of llm and um, geni and stuff I love to see that, getting more and more mainstream as well, one of the things I'm always thinking, about is U skill is one thing because we, a lot of the applications that we talk, about today especially in the context of, geni we always talk about like the, bigger computer and bigger scale I would, love to see that getting smaller which, it is happening now getting more, accessible on different devices and, stuff being able to do more cool stuff, on device and band stuff on they excited, for that too yeah yeah I think there's a, lot of people excited for that and sort, of this new phase of AI where people, talk about AI everywhere or this sort of, thing which in reality you know there's, been machine learning and data science, sort of everywhere for for some time but, that that sort of wave of these newer, generation of models kind of being, runable in in more practical scenarios, is is exciting but um yeah thanks for, thanks for joining beanson to to talk, about a little bit of your broccoli AI, it's been uh it's been fun and uh love, it thanks for indulging me yeah yeah you, you and I can can hype the broccoli Ai, and I'm sure we can get Demetrios uh to, help us hype it too yeah I don't know if, we trademarked that term uh he's got it, in in his hype cycle now, so yeah yeah thank thanks so much for, joining and hope to talk to you again, soon thanks thanks for, [Music], every all right that is practical AI for, this week subscribe now if you haven't, already head to practical a.m for all, the ways and join our free slack team, where you can hang out with Daniel Chris, and the entire Chang log Community sign, up today at practical ai. fm/ Community, thanks again to our partners at fly.io, to our beat freaking residence, breakmaster cylinder and to you for, listening we appreciate you spending, time with us that's all for now we'll, talk to you again next time, [Music], K LOVE |
UCZbtcRQHGpU_tt2yM4C7k0A | NRrH9-kRMwc | Changelog | youtube#video | Hyperventilating over the Gartner AI Hype Cycle | This week Daniel & Chris hang with repeat guest and good friend Demetrios Brinkmann of the MLOps Community. Together they review, debate, and poke fun at the 2024 Gartner Hype Cycle chart for Artificial Intelligence. You are invited to join them in this light-hearted fun conversation about the state of hype in artificial intelligence.
Leave us a comment (https://changelog.com/practicalai/279/discuss)
Changelog++ (https://changelog.com/++) members save 5 minutes on this episode because they made the ads disappear. Join today!
Sponsors:
• Intel Innovation 2024 (https://intel.com/innovation?regcode=CMCCHL&utm_campaign=Changelog) – Early bird registration is now open for Intel Innovation 2024 in San Jose, CA! Learn more (https://intel.com/innovation?regcode=CMCCHL&utm_campaign=Changelog) OR register (https://reg.oneventseries.intel.com/flow/intel/innv2024/InnovationReg?regcode=CMCCHL&utm_campaign=Changelog)
• Motific (https://www.motific.ai/) – Accelerate your GenAI adoption journey. Rapidly deliver trustworthy GenAI assistants. Learn more at motific.ai (https://www.motific.ai/)
Featuring:
• Demetrios Brinkmann – Twitter (https://twitter.com/Dpbrinkm)
• Chris Benson – Twitter (https://twitter.com/chrisbenson) , GitHub (https://github.com/chrisbenson) , LinkedIn (https://www.linkedin.com/in/chrisbenson) , Website (https://chrisbenson.com)
• Daniel Whitenack – Twitter (https://twitter.com/dwhitena) , GitHub (https://github.com/dwhitena) , Website (https://www.datadan.io/)
Show Notes:
• MLOps Community (https://mlops.community)
• MLOps Community Podcast (https://tr.ee/2aUfMm9AIb)
• Gartner Hype Cycle for Artificial Intelligence, 2024 (https://www.gartner.com/en/documents/5505695)
Something missing or broken? PRs welcome! (https://github.com/thechangelog/show-notes/blob/master/practicalai/practical-ai-279.md) | 488 | 5 | 1 | [Music], welcome to practical AI if you work in, artificial intelligence aspire to or are, curious how AI related Tech is changing, the world this is the show for you thank, you to our partners at fly.io fly, transforms containers into microv VMS, that run on their Hardware in 30 plus, regions on six continents so you can, launch your app near your users learn, more at, Y.O what's up friends Intel Innovation, 2024 is right around the corner, accelerate the future registration is, now open and it takes place September, 24th and 25th in San Jose California, this event is all about you the, developer the community and the critical, role you play in tackling the toughest, challenges across the industry ignite, your passion for AI and Beyond grow your, skills to m maximize your impact and, network with your peers as they unleash, the next wave of advancements in, technology here's what you can expect, understand the emerging Innovation and, Trends in Dev tools languages Frameworks, and Technologies in Ai and Beyond to, empower you and the solutions You're, Building get in-depth technical, experience join handson workshops Labs, meetups and hackathons to collaborate, and solve problems in real time you can, explore featured partner and Intel, solutions they have Partners there, startups there customers there and Intel, is showcasing the latest in products, services and solutions across Keynotes, Tech sessions and the show floor to help, you meet your development needs, collaborate with experts learn and have, fun engage in interactive sessions to, connect get certified gain unique ideas, and perspectives build longl lasting, networks and of course have fun and get, inspired hear from leading industry, experts technologist, startup entrepreneurs and fellow, developers along with Intel leadership, CEO Pat Ginger and CTO Greg lavender as, they take you through the latest, advancements in technology don't miss, this chance to be at the Forefront of, innovation take advantage of early bird, pricing right now until August 2nd, register using the link in our show, notes or to learn more go to, intel.com inovation once more that's, intel.com Innovation or go to the show, notes and click that link, [Music], well welcome to another episode of, practical AI this is Daniel whack I am, founder and CEO at prediction guard I'm, joined as always by my co-host Chris, Benson who is a principal AI research, engineer at locked Martin how you doing, Chris I'm doing fine uh we got a fun one, today Daniel just give you a good one, yes of course it was wonderful not that, long ago to be in uh the great city of, San Francisco and run into our friend uh, Demetrios from the mlops community and, uh I I figured I'd just bring him along, for another conversation so Demetrios, how you doing I'm great man we're back, and I've got some bad news to break to, you right now I wanted to do it on air, go for it just to get your reaction oh, boy you can be vulnerable this is how we, build community yeah I'm nervous yeah so, prediction guard awesome congratulations, on all the success that you've had we're, doing a data engineering for ML an AI, virtual conference and one of your, colleagues Daniel filled out the cfp I, haven't gotten back to him yet but I, can't accept him I just am way too full, way over my head and as much as I want, to I'm going to have to divert him to, doing his own special event basically, we're going to actually take what may, have been a bad thing and turn it into a, good thing that that sounds great I'm, looking forward to learning more there, we go you know I gotta I got to make, sure that you get all the love and shine, you deserve because I'm super stoked at, what you're doing yeah yeah we, appreciate that it was great to see you, and and you had your own event in SS how, was that I do not recommend doing Live, Events to even my greatest enemies if, anyone out there is contemplating, organizing an AI conference you can do, it but I don't recommend it you're gonna, hurt yeah painful man but it it was a, big success it was just a lot of work, leading up to it as you can imagine and, we had fun and on the day of it was like, I think over 750 people showed up, a lot of great conversations a lot of, fun like spontaneous sporadic meetings, with people and that's the stuff you get, at in-person conferences that you it's, really hard to replicate virtually yeah, you know what the secret is the secret, is it's Ai and it needs a lot of hype it, really needs a lot of hype there's one, thing that we don't have enough of in AI, it's we don't have enough hype if you, had hyped it more it would have, worked you know I do a fair amount of, hyping and so for those out there that, are sick of the hype like myself I've, only got myself to blame on this well uh, Chris you sent me um a very interesting, looking hype filled chart the other day, you want to go into what that was well I, I will uh and I'm actually blaming it, all on Demetrius uh he was making fun of, the G hype cycle and gosh I hope they're, not a sponsor because we're making fun, of them today and and he was he was, going through that on and it was funny, and I said dude we need to do an episode, where we all analyze the Gartner hype, cycle in 2024 for artificial, intelligence and we we break it down and, we're going to assess it and decide what, we think of those things and we're we're, not doing this in our normal extremely, serious manner we are doing this in the, fun way and and un L you don't know, Demetrius out there which I can't, imagine because he's a regular guest on, the show here he is in addition to being, a brilliant guy in this field he's also, the funniest man in all of artificial, intelligence so this is going to be good, uh and we're going to dive into the, gardener Hive cycle today and break it, down for you we're going to start with, the real one and then we're going to uh, maybe make some adjustments to it you, know Chris you you say making fun but um, I mean Gartner seems to have fulfilled, their mission I mean we're talking about, the the hype cycle we're we're going, into it so maybe their mission was, fulfilled and you know we are their, fulfillment yeah oh my gosh yeah we're, hyping it up right now we're H very true, okay and we're gonna have fun doing it, oh I just have to say yeah please if, anyone knows how I can get a job doing, this kind of stuff just making up words, and then putting them onto a wave graph, let me know because I would love this as, a job it just seems like it's too much, fun well let's see I think Surf's Up let, top on the wave and let's start talking, our way through you know Demetrius you, wanna do you want to lead off on what, some of your ideas there so I think the, most surprising to me out of this whole, graph and for anybody that's not, familiar with the hype cycle you've got, the big like upward side and then it, goes down and it kind of crashes and, then it starts to climb back up and it's, the traditional like and the two second, version of that and I did a in our in a, previous episode I did a longer version, uh when we were looking at some specific, things on it but the two second version, is new technology comes out everyone's, super excited about it they think it's, going to be the greatest thing since, sliced bread uh it doesn't live up to, the hype they get frustrated they go, good this thing sucks and and it falls, down on the hyp popularity side and then, cooler heads prevail and they kind of go, okay well maybe it can do something okay, and and then it's into a a reasonable, sense of productivity so that's Gartner, in a nutshell so the biggest surprise, for me is at the bottom of the slope so, after it's gone all the way up the hype, cycle it's come down and crashed down, and it is at the absolute bottom the, trash of, disillusionment exactly there is cloud, AI service, yes and for me that is the biggest, misnomer because if anybody is making, any money out of any of this and I guess, maybe hype and actual money they're, detached and they're very decoupled here, but for me that was like wait what, there's no hype in Cloud AI services so, Bedrock out of there hype is killed it's, at the trough of disillusionment any, type of sage maker if you're using that, or vertex no out of there it's the, lowest of the low and so when I saw that, that was instantly like dude why are you, even doing it yeah I did not believe a, thing that I read afterwards but that, was my thing any any big surprises from, you guys I think you're point on if, there's anyone making a killer amount of, money on this it's Microsoft it's Amazon, it's Google uhhuh part of my struggle, here is some of these terms like I could, interpret them one way or another way, right like Sage maker for example which, for those that don't know is a it's kind, of like a model deployment service, within AWS and there's various, convenience around it and that sort of, thing like that's been around for quite, a while now like a very long time even, before sort of the kind of piped gen AI, stuff long for it but yeah so like is, that a cloud AI service like that's been, around for a huge amount of time or are, we just talking about like hosted model, apis right they don't say which also to, be fair have been around a long time, like you look at something like OCR or, translation or something like that and, and cloud services have been around for, a really long time and are sort of, ubiquitously used it's funny that it's, down there I I I get your point maybe, it's just like everyone knows that's, where the you know Cloud that's where, all the services are we're all paying, for them yeah so does hype correspond on, to usage I guess like in this chart is, it that people aren't hyping Cloud AI, Services even if they're used or I I, think it's an emotional thing you know, the hype side is you know how much, people are talk so maybe it's accurate, in this context there is nothing sexy, about AI services in Cloud providers and, maybe that's what they're getting at is, like yes we're paying an arm and a leg, we're giving them all of our money but, there is nothing sexy but productivity, wise it's definitely productive I I, would think so yeah it's very pragmatic, too especially for those people just, starting I don't know any easier way, than to just grab an API from like, Amazon Bedrock is just hosted model hit, that API like you would hit an open AI, API but now you have a suite of models, right so that seems to me like a a near, Miss but then at the top of the peak is, the other one that was a huge surprise, to me because because I've noticed this, trend I don't know if you guys have, noticed it but people who were formerly, ml Engineers we've all converted into, being AI engineers and an AI engineer is, so misleading because you don't know is, that somebody that is coming from like a, front-end development world and now they, do a little prompt engineering they use, a few Frameworks and they can chain, together some prompts to make a bit of a, demo on Twitter and now they're an AI, engineer or is it somebody that was deep, deep in the ml platform weeds and, because AI is now the new rage they call, themselves an AI engineer so I I don't, know about that but it's at the top I, think it's the same yeah I think it's, all I I think people use AI ML and, before it really fell out of Vogue deep, learning, interchangeably yeah so exactly I don't, know if it's also maybe connected to the, fact like Chris and I talked about this, I believe it was maybe last week the, fact that some of the disillusionment, around AI is sort of the realization, that turns out AI is integrated in, software and you still have to do, engineering to like build software and, it doesn't just sort of like having a, model is a solution doesn't really like, play out in reality you mean I can't, just buy an AI model and stick it out, there and magic things happen yeah I I, mean one would think I'm so, disillusioned yeah it's it's funny you, guys mention that too because I've seen, a few people talking about how llms are, not a product you have to build on top, of llms your product or whatever it is, your service that needs to be there so, you can't look at an llm as a product, per se and then I've also seen or I've, been thinking deeply about something, that is like the companies that are, really getting a ton of value out of, this AI movement uh I'm thinking about, one of my friend's companies who does, like a support software and now he's, leveraging Ai and llms for creating like, multi-agents and helping answer feedback, or answer questions and queries for, support and he's using AI that's awesome, he's able to sell that support product, two companies really well what I haven't, seen is companies that say Hey I am, fraud detection as a service and I'm, going to sell you this whatever, traditional ml product as a service, whereas you can create regular business, unit products as a service that leverage, AI but you can't quite or at least I, haven't seen anybody crack the nut, create some kind of a traditional ml, service type of product I don't know if, you guys have seen that and I also don't, know if I'm making much sense right now, because it's something that's relatively, fresh in my mind I'm gonna turn that one, over to, Daniel so no I wasn't making much sense, I guess is what the nice way of saying, it is um I mean so you've got like what, I would say is the things that I have, seen most are either what you were, talking about so, utilizing generative AI embedded in the, functionality of sort of domain specific, applications like the customer service, you're talking about or financial, services or whatever or access to models, over some API infrastructure right, there's maybe less like General I I, guess maybe the biggest one I've seen is, sort of just general like fine-tuning as, a service if you look at something like, you know open pipe or or something like, that but that's still fairly general, purpose it's not specific to any sort of, use case that you might use maybe to, some degree you know certain rag, Services would fit into that like we, talking to Pine Cone about their recent, like they have more kind of pre-built, things to have you do kind of like load, in all your documents and have rag set, up and all all that stuff so um I don't, know that's maybe the closest that I've, seen to to that sort of, scenario yeah well also the big question, is everybody wants to and this kind of, ties back into the hype cycle everybody, wants to be doing Rag and wants to have, all these great use cases with their Rag, and so like you were talking about with, pine cone they make it really easy for, you to do your rag but then at the end, of the day is that a viable business or, is that actually super useful as opposed, to somebody's got this support software, that they can come in and really cut, down the burden for your customer, success Engineers or your customer, success people and that is fascinating, to me because it's it's a booming, business right now the rag business, maybe yeah that's great maybe there's, some interest there is it a booming, business I don't know I haven't seen, numbers but I think the really, fascinating part to me is if you try to, juxtapose that with like a fraud, detection as a service type of product I, just haven't seen that anywhere because, I think a you're not able to really like, give away everything as freely and be, what works for one fraud detection use, case doesn't necessar it's not like you, can productize that and then go out and, sell it as a service and my opinion so, so this is a little bit of a tangent I, know but uh but that all that to say is, we're at Peak hype for AI, Engineers Peak hype yes so I'm gonna, draw us back over to the hype cycle just, for a moment and I want to read I'm, going to do something boring for a, moment I'm going to read off the things, where they are uh for our listeners, because the three of us have the benefit, obviously of seeing the graph in front, of us and for listeners who aren't so, I'm going to take a a moment and then we, can go back and start hitting them there, very quickly heading up the curve, initially The Innovation trigger we have, autonomic systems we have Quantum AI we, have first principles AI we have, embodied AI multi-agent systems AI, simulation causal ai ai ready data, decision intelligence neuros symbolic AI, composite AI artificial general, intelligence otherwise known as AGI and, then we're hitting the peak of inflated, expectations at the top of that hype, cycle we have Sovereign ai ai trism, prompt engineering responsible Ai and at, the very Peak AI engineering and then, starting to slide down we have EDI, Foundation models synthetic data model, Ops and generative Ai and just going, into the trough of disillusionment is, neuromorphic Computing smart robots, followed at the bottom by Cloud AI, services and then we slide up the slope, of enlightenment to autonomous vehicles, knowledge cfts intelligent applications, and finally the singular one on the, plateau of productivity which is where, you want to end up is computer vision, which is basically yeah we can do that, it's boring and no one talks about it, anymore but hey we're making money so if, the listeners out there are not confused, oh there's a whole bunch I don't have, any idea what they, are that's it I was going to say which, ones do you actually know what they are, because what the hell is embodied AI oh, I I learned what that is after I put out, the post so someone said oh yeah, embodied AI is when you use AI in robots, it is so yeah but there's also smart, robots on the on the cycle and I used at, a former employer I was specifically, doing AI systems in robots and I've, never heard of it you never called it, embodied AI well it's been a few years I, I'll give you that it was so but no we, weren't calling it, I mean so I think I'm at like a 30% hit, rate on these and I really would love to, know what first principles AI is because, that feels like buzzword Bingo to the, fullest I don't know um let's see first, yeah Daniel's going AI he's going to, models to find out he's um the card AI, generated card in my Google search says, when applied to AI first principles AI, suggests developing AI systems and, algorithms by understanding the, foundational principles of machine, learning neural networks and data, science from the ground up don't we do, that anyway when we're isn't that kind, of inherent in training new models and, stuff like oh but no no we're really, going back we're going back to the very, first ones you're at the second or third, principal we're beating you yeah no cuz, all you guys that are out there that, aren't using first principles you know, that's lower down on the hype cycle okay, this is, yeah so the other pieces I I mean were, there any other surprises for you guys, because I have so many other pieces on, here that I'm like what I think for me, like some of these things are themselves, correlated and yet in different places, on the chart right so it's like if you, look at generative AI Foundation models, Edge ai ai engineering prompt, engineering probably some others on, there all of those like sort of fit into, the sameish bucket and yet are on, different sides of the hump so yeah I I, don't know like some of these it's also, a matter of where do you draw the, boundaries where's the boundary between, generative Ai and Foundation models or, generative Ai and prompt engineering, I'll give you one you know as where at, the very bottom on the Innovation, trigger is quantum Ai and I've okay so, that's not gonna happen anytime soon and, and I will note that they have it only, greater than 10 years but I would, suggest it's probably greater than, greater than 10 years but isn't that uh, I mean one of the things that's, interesting about this whole cycle is, there's that one uh maybe you all can, tell me or I can look it up there's a, one law it's like a general law that, people talk about where you, underestimate short-term Innovation and, overestimate longterm Innovation or or, something, vers yeah yeah sorry I said that, backwards yeah so it seems like some of, like it's hard to especially the time, angle of this it's hard to because, things just pop up and you like really, didn't see certain things coming and, others that you thought would would come, don't so yeah it's extremely difficult, 100% one thing that I am just to tag on, what you're talking about Daniel with, the bucketing these, please tell me what the difference is, between an AI engineer and a prompt, engineer what like a prompt engineer is, someone that only does prompts I guess, and that's all that matters so they're, just so I can see how how it's like, where's the line here when prompt, engineering came out Daniel you might, remember I kind of made fun of that I, was like the whole that you talk about, like because people were saying they're, new jobs for prompt engineers and stuff, and I'm like that is a passing fact bad, like that will be just so ingrained in, what everybody does all the time that, the notion of there being someone who, that's their entire job all the time for, years is not gonna happen yeah I also um, didn't know so like I've never heard, anyone use the word or if it's a word, it's an acronym AI, trism do people go around saying that, yeah what is that what is it so it's I I, looked it up and you know what's what's, funny because this is exactly the area, that I'm working in every day it's AI, trism is tackling trust risk and, Security in AI models ah okay you've, never heard that used have you and I've, never heard that but now I feel like I, should put it on our, website because it's hyped yeah you, definitely need there that's right the, funny part is it's almost as hyped as, prompt engineering which you is, basically all you hear about is prompt, engineering right and yeah they're right, there together AI trism you never hear, about yeah there you go uh but the trism, it's it's out there it, is we hear about you know the the, components that make that up all the, time sure but just never the I've never, heard them put it together that way and, I'm sure there are people that are out, there that that you know their focus is, in the that area and they're like of, course it's trism how do you but yet, guess what most of us don't know that no, not at all I don't even know if I go and, I just look at this I don't know what, causal AI is I don't know what the AI, simulation is the multi-agent I do, understand but, then like even when you say Quantum AI I, don't know what that is the one that I I, would say is, probably in the wrong spot is synthetic, data it feels like that should be still, going up on the hype train because we're, just discovering what we can do with, synthetic data and every week I feel, like we unlock new use cases and, synthetic data is just a it's the gift, that keeps on giving in my eyes I think, that's the difference in you who, actually does it and somebody at, Gardener who you know who was tasked to, go put the chart together and doesn't, actually do the thing in real life I've, terribly offended somebody out, there well we're glad that it's out, there let's just say that we are very, happy that this exists so we can have a, whole episode dedicated to breaking it, down yes it's a conversation starter, that's what I mean like, achievement unlocked yeah unlocked so, one thing that I noticed isn't there at, all which really surprises me uh given, how much it's bantered about is ethical, AI it's not on the chart and that does, doesn't go in the trism that's not one, maybe it does maybe this is where I you, know is ethical AI now transformed from, a labeling standpoint into trism is that, is that where we're going I don't know, or what is the overlap between, responsible Ai trism and ethical AI okay, well yeah I don't and there isn't really, anything on here about gpus or Hardware, so yeah I think that's because they made, their own hype cycle for, gpus right if I'm not mistaken I I feel, like I've seen that somewhere on the, internet you'd be cannibalizing your, other chart exactly so you can't put any, GPU Hardware anything on the AI one you, got to refer people to the GPU hype, cycle and maybe it's like that with, ethical AI like they made a whole other, ethical AI chart that is the hype cycle, for ethical AI maybe so I'm not familiar, with it how many charts can you make, that's if you're Gardener I guess think, they have I mean we have just the, artificial intelligence hype cycle here, but they probably have I think I've seen, multiple you know subdivisions and stuff, out there so that's why it's a great, business to be in Gardener selling all, these different hype Cycles well, speaking of what to Hype what uh what's, not on the hype cycle but should be all, right if I could have talk to somebody a, gardener before they were making this I, would have advised and so this is my, basically this is my video job interview, right now I'm busy typing an invoice up, for you to send to them okay just for, exactly I would have advised AI Gateway, that is very popular that's climbing the, hype cycle right now because people, really like to have the option to hit an, AI Gateway and if it is not that compx, of a query you don't need to hit gp4 you, don't need the most expensive model if, you have some kind of Open Source model, that is cheap then let the simple query, go to that 7B model and so I've been, hearing people call it an AI Gateway, others I think have called it like a llm, proxy maybe or router yeah that's, another one so we would have to agree on, the actual name but that's gaining hype, for sure yeah agreed yeah it's uh I've, definitely seen the router language, whatever it is like the languages, overlap with networking um which is, basically like you're just routing API, calls so I guess that makes sense yeah, any any that you guys would have liked, to have seen on here and where I had the, ethical I'm still wondering what, composite AI is did we ever get that, answered or if I just am I having a, senior moment what is it yeah what is it, um what uh the one that really stands, out to me unless I'm just like there's a, lot of words on this page so maybe I'm, totally missing it somewhere but where, is multimodal AI yeah oh good catch, there it's not on here is it no who, cares about multie that's so weird that, should be in the peak of inflated, expectations this is like the thing of, 2024 like multimodal AI That's so even, multimodal rag should be on here like, climbing the Innovation trigger, multimodal models should be on the peak, of inflated expectations that that is, such a good catch I know tons of people, who who say multimodal and have no idea, what it means well what what does it, mean Chris yeah, what quiz time well let's having, different modalities of of input there, so that you can combine different inputs, to get a a rich output you know in a, very general sense I have no idea yeah, so voice I know when I photos yeah, photos videos all the things all the, things exactly which is what we want I, want to throw a bunch of stuff that I, have and and have a fantasttic just have, it sorted out and give me the best, answer uh and even with today's, multimodal models that doesn't happen, very well there's I I'm I'm often I'm, often frustrated and disappointed with, uh with those outputs so yeah it's I I'm, expecting better yeah and along those, lines I have two that I would like to, have seen one is just transformers in, general whereas that where are they on, this hype cycle because that also feels, like are they climbing or are they going, down I don't know it would be trough of, disillusionment heading downward because, that it's kind of we're we're we're past, that and people are now talking about, post Transformer models you know quite, often so it's kind of like yeah, yesterday so there needs to be another, dot for post Transformer models going, that's definitely going up and that's, right speaking of which it feels like, okay we've, got small language models where are they, because that is all the rage it is that, was like and maybe it's all the rage for, every vendor who is not open AI because, they can't compete on gp4 and so what do, they do they say well you can just host, your own small language model and, fine-tune it and get better performance, than, gp4 and, so I think small language models are, probably they should be in that, Innovation trigger maybe the peak of, inflated, expectations because anyone who's ever, used the 7B model, might not want to use it if they have, the choice well maybe maybe it's is it, are you sure that's going up or could it, possibly be sliding into that, disillusionment that you just referred, to potentially that's true because maybe, it is going into the trough of, disillusionment uh you know hypo just, hypothetically because I do think that, when it gets to the plateau of, productivity small models will be this, the just the Workhorse you know you'll, have them out on the edge everywhere, every freaking device you've ever, imagined or seen is going to have small, models in it that are inferencing we, won't ever have anything that doesn't, have them it'll be just the oon of, course we have our small models in our, watch which leads me to the next one, that I'm like where is this why do they, not have wearable AI that is a perfect, buzzword that should be on here and if, you look at like what meta is doing with, the glasses or if you see any of those, necklaces that you can wear and it, records everything yeah that's wearable, AI right there I just I may have just, made that up or I may have seen that, before but that one should be on here it, should be there I, [Music], agree hey friends out shift Cisco's, incubation engine merges Innovation with, the art of possible a Launchpad for, transformative emerging Tech out shift, Blends startup agility with corporate, strength to develop nextg Technologies, from the groundup in AI Quantum, Technologies Cloud native and more their, newest AI Innovation Motif addresses a, critical challenge in the rapidly, advancing world of gen AI Bridging the, Gap between concept and deployment this, model and vendor agnostic solution, supports the entire gen AI Journey from, assessment and experimentation Motif, accelerates deployment from months to, days while safeguarding against gen AI, security trust compliance and cost risks, all while empowering business function, and it teams to rapidly configure and, user assistance powered by, organizational data Motif provides, Advanced customizable policy controls to, prevent unauthorized access to sensitive, data and helps ensure compliance, throughout the entire process with deep, visibility into operational and business, metrics motivic enables you to track Roi, optimize costs and make informed, Decisions by offering a centralized view, moic deters Shadow AI usage and empowers, teams to innovate responsibly so move, beyond the traditional constraints of AI, implementation utilizing AI deployment, that is both responsible and is, revolutionary ensuring your projects are, not just just quickly launched but built, on a foundation of trust and efficiency, visit motif. a that is m o t f i c., [Music], AI maybe this fits into kind of the, agentic stuff that is repres presented, in certain ways on there but this whole, idea of whatever you know like tool, function calling SL like text, tosql like interacting with structured, databases apis whatever that is I don't, know like the maybe the general name for, that other than tool and function, calling or text to SQL but um certainly, that's like sliding into a Zone where, people are definitely doing some of, those things in production and there's, products released around it so like the, the hex magic stuff and and all that, that other where is it on the chart, though before I go on oh where is it on, the chart um I mean it's got to be, somewhere somewhere around AI, engineering so it's at the peak of ex uh, maybe maybe yeah I don't know I don't, know it might be maybe it's going down, cuz people are like H agents aren't, reliable I think that's right I think, it's heading down into the TR, disillusionment that's where I would, guess yeah Y and if you compare that to, where they have it multi-agent systems, it's got a long way to go up it is at, the very bottom of this hype cycle so, yeah I think, we instinctively are like no please no, more agents and Gardner's like oh we're, just getting started baby well and and, they're like no please more agents, together multi- agents it Gardner's, going create their own agent Hye cycle, next that's going to be the next one, that they can create and so we'll you, know take a commission for giving you, that idea Garder no problem there and, one thing can we call out the elephant, in the room because where is retrieval, augmented generation, on yeah how is that not on here really, yeah rag what's that cuz I was thinking, about it I was thinking about it and I, was like oh you know what they missed is, graph rag that is all the hype these, days and that's probably right around, where Sovereign AI is where it's maybe, like at the border the yeah it's going, up nearing the peak of exp of inflated, expectations you're right more hype than, the trism yep more hype than the trism, but I would argue rag is is heading to, the trough of disillusionment anyone, want to disagree with that no no I think, so too I think it's over the hump yeah I, do too I mean it's and people are kind, of hitting the the challenges and and, and you know and actually Daniel, Advanced rag you know which we've talked, about several times you know kind of, kind of trailer well we don't just have, rag now we have advanced Rag and and, advanced as things are starting to head, over that peak of inflated expectations, with rag well guess what we can juice, some more we have advanced rag but I, think I think the whole thing is, starting to go over the side you know, people are like okay well we've kind of, done at least the easy stuff uh to the, advanced rag point there are people that, are that are doing it better than others, but nonetheless you know it's you know, what's next so what what I'm just, curious to Second deviation we've talked, about you know fine-tuning we've talked, about rag what's coming next in that in, that sphere what what are they missing, there yeah a new model yeah I think you, mentioned that you might have had some, of these Demetrios uh what are AI hyped, items that are your own that you've come, up with a name for oh that other people, will have to, interpret to to figure out their, definition you wanna you want to guess, yes on this one all right here we go I, am going to start you off with a with a, pretty simple one this one is free range, AI free range is that is that open, access, llms close close what do you got Chris, grain fed I I can't get off the free, range thing I'm an animal guy I can't, even I can't even get into the AI head, space on this one it's AI that was, trained without guard rails okay gotcha, okay I like that well we we already, talked about about one here um that that, you alluded to Demetrios but my name for, it was trinket, AI, wearables yeah yes trinket AI TR yeah, imagine it's it's in your fidget spinner, that sounds a lot that's a much better, name than wearable AI yeah trinket, AI it is every little thing you have on, your body has a freaking model INF, single in it you know, you're and it doesn't bring you any, extra value if we're going to follow the, AI Trend you just don't have to think, anymore you can click that button and, take a picture Demetrios no it just, gives you some verbose answer to a, question that you didn't really, ask so your shirt is you're like hey, have I been sweating and then it tells, you the origin of sweat in a three-page, PDF that you have to go, download well do I get senior moment AI, that that would be good for me you know, I there's a huge market for that every, everybody over the age of you know 50 is, going to buy senior moment AI to to you, know like what what what oh and it oh, there we go and you know I can I can, continue instead of pausing for the next, three minutes to try to figure out what, it was I was about to do or I was, thinking that that's a how seniors, interface with AI so they don't get left, behind it's like this is the product, that will make sure you stay up to date, you're ahead of the, curve okay, sounds good all right I got another one, for you all this one is, EQ AI oh empathetic AI yeah so it's also, been known as empathetic AI yeah you may, hear other people out there in on the, streets calling it empathetic AI uh this, one is a type of AI that has high, emotional intelligence and it feels, empathy for you when you get frustrated, that it's not giving you the right, answer and your prompts aren't working, but it doesn't actually make your, prompts work it just feels bad for you, okay I that minus the AI bit that, happened to me yesterday I was on, Comcast on their stupid text support for, four hours texting they passed me off, and every everyone was so empathetic but, they accomplished nothing if you put, that in AI I'm quitting AI if you put, that into any AI that does that I'm just, done I'm I'm walking away from the whole, field are you sure it wasn't already AI, that you were talking it could have been, I mean it was just text it was only text, but it was horrible we've already passed, the Turing test so it's like they're, there I'm getting a response of I'm so, sorry I'm just very sorry we're here to, help you and I'm like I'm gonna freaking, kill you you know yeah yes that's what, four hours texting support will do but, don't do yeah I just I'm if you bring, that to AI it'll ruin the whole thing, for me, well this one funny enough is actually, on the up Tick when you look at the, slope the EQ a has got a lot of Runway, left yep um so my my next one is, AI either AI nepotism or AI, anti-nepotism, oh don't I'm trying to make fighting, fighting AI nepotism fighting AI, nepotism oh okay you're gonna have to, you're gonna have to go into that one, for me that's I've stoed you yeah yeah, yeah this is, exciting it's, basically using AI against like the, government using AI or what no no so uh, uh Foundation model related maybe yeah, so this would be like, multimodel AI in that you are not, preferential to one language model, family and only using that family but, you are now multimodel and you know as, such not practicing nepotism but are you, multimodal, multimodel this is also you know I knew, it by its other name uh which is, polygamy, AI yes oh gosh where are we, going ohy no or or some in San Francisco, call it polyamorous AI as it tends to be, so the the next one that I've got for, you oh where is this nepotism AI on the, hype cycle by the way uh I think it's, still a bit on the rise I saw a16z in, their in their post one of the things, they called out was multimodel future oh, yeah there's a future for this one that, is for, sure so I've got one that is called, broccoli AI okay this one's on this, one's going down is it related to some, sort of graph thing no but that could be, nice yeah branching is it synonymous, with healthy AI yeah yeah exactly maybe, you've heard it termed healthy AI, efficient yeah it's sustainable no so oh, that's another one that I've got though, but we'll get to that in a minute which, which reminds me like it does feel like, like sustainable AI should have been on, the real hype cycle like that's an, actual ter isn't it yes it is and it's, not and it's not on there the other one, that should have been on there that I, was like why isn't it on there is, Ensemble AI that feels like or Ensemble, models that feels like it should have, been on there see one of the ones that I, looked up was composite AI yeah that's, the one I didn't know I think well I, don't know it's slightly different than, Ensemble but I think that they like, composite was combining multiple, multiple AIS together um in some way or, another for one inference like you have, multiple you know models inferencing but, you have one inference back out to the, uh the user yeah something like that I, don't know although Ensemble could very, yeah very much mean for a single, inference getting a majority vote or, something like that okay so it would be, where composite AI is on the chart if, they're assuming they're correct yeah, and where where before we leave it, sustainable AI where is it on the chart, that's very much like it's got a lot of, hype to go because mid level mid level, on the on the curve up yeah okay just, think about how many people are talking, about the energy that is wasted training, the foundational models true and how we, need to build out all these data centers, and they need to be sustainable etc etc, so yeah sustainable AI for sure has some, room to grow back to BR, AI AKA healthy AI this is AI and this is, very much on the Downs slope again it, has passed its peak people are a little, disillusioned with it because it's AI, that doesn't taste good for the, organization but it's needed and so you, can imagine the cyber security folks, they love this kind of AI is this like a, linear aggression model or what would, you consider good for an organization I, think you use the word good yeah healthy, it's it's healthy for the we could go to, healthy for the organization what could, that be I I mean I actually didn't get, to do enough market research in this, section, to to figure that part out you know I, was just throwing spaghetti at the wall, but I if I were to think about what's, healthy yeah it would probably be the, traditional ml going back to the what I, was talking about before like fraud, detection is one of those, where it's not really AI some people, might know it as its former term ml, so I'm telling you they're all the same, from a marketing standpoint exactly well, yeah the waters are too muddied for them, to make any actual difference that's, right so what else you got what else you, got okay so I've got unsustainable AI, which is way different than sustainable, AI just so we're clear an inverse, but it's not even it's a whole, different uh sector of the universe that, we're talking about it's not like oh, it's just the opposite of sustainable AI, unsustainable AI is it's got It's at, pype right now let's be honest if I, could swap it out with the AI engineer, it is at Peak hype because this is AI, that was built for a product demo but, not for scale that is unsustainable AI, happens all the time yeah so anything, that you see um basically we can, hopefully none of these guys are your, sponsors but let's just cue Devon or, rabbit or Humane all those unsustainable, AI the trinkets the trinkets yeah that's, true so it's sort of analogous to doing, like prototyping software where you're, you're never intending to to grow into, production exactly so so that's all of, mine that I I could think of well I, think that was a pretty good list I did, realize I don't know maybe maybe related, to some of the discussion we had earlier, but I don't see neighborly AI on here, that's kind of creepy when you think, about it I I wasn't creeped out until, you said that, but I had this image of Mr Rogers, Neighborhood you know but instead of Mr, Rogers it's the AI hi girls and boys, maybe they can, help you clean up a few things with, their Rags it clean, no oh boy oh well I was thinking it was, like next door where it was almost like, the voting system The Ensemble but it, was for local llm gotcha yeah I realized, there's nothing about vectors or, embeddings on the chart I was just, thinking about that actually there's, yeah there's no Vector stores on here or, even just general embeddings, yeah I wouldn't that be plateau of, productivity now we've had those for so, long that they're just I don't know, lexicon no no emotion left in them yeah, what I was thinking is they probably, aren't on there because Gardner also has, one of their best products ever the, magic quadrant and that'll be the next, episode that I come and drop in on we, can remake the magic quadrant for the, different sectors and I imagine, that they have a magic quadrant for, Vector databases yes that sounds, delightful yeah well it it it has been, delightful to uh to have you on, Demetrios um I'm glad you brought your, various new AI terms to the hype cycle, and uh now I have have some work to to, do on my broccoli AI so incorporate that, into your product for sure it's it's, right around there prism it would be a, good AI logo just like a broccoli fluet, yeah the broccoli or the I saw a great, paper that was all about leaks it was, all about data leakage when you send API, calls to open Ai, and the paper started with a emoji of a, leak that's awesome like the leaks you, eat right and it was saying here's and, it was basically showing how you send, your data to open AI but but a lot of, other people are going to get it too if, you're not careful yeah which is one one, thing that we haven't really touched on, but that seems like it's got some hype, around it is what data leakage AI data, leakage data poisoning data poison I, know in my in in my day job that's a, common conversation yeah prompt, injection should be there injection yes, uh I guess this all fits under trism, yeah this is a we're going over trisms, right now trisms and, trinkets on that note that very profound, note uh it has been great to discuss the, all the trisms with you uh Demetrios, I've had a blast as always please please, come back uh as as usual give your uh, your own hype about the the upcoming, event before we close out and where, people can find out more about it yeah I, I always feel bad I come on here and, just show my stuff so this time no, Shilling I just had a blast doing this, with you guys okay so if anybody wants, to find out about the next virtual, conference or the inperson conference, they can just Google mlops community and, I'm sure it'll pop up cool all right hey, much appreciated we'll talk to you soon, demitros thanks man yeah thanks, [Music], guys all right that is practical AI for, this week subscribe now if you haven't, already head to practical AI FM for all, the ways and join our free slack team, where you can hang out with Daniel Chris, and the entire change log Community sign, up today at practical ai. fm/ Community, thanks again to our partners at fly.io, to our beat freaking residence break, master cylinder and to you for listening, we appreciate you spending time with us, that's all for now we'll talk to you, again next, [Music], time k |
UCZbtcRQHGpU_tt2yM4C7k0A | NRrH9-kRMwc | Changelog | youtube#video | The first real-time voice assistant | In the midst of the demos & discussion about OpenAI’s GPT-4o voice assistant, Kyutai (https://kyutai.org/) swooped in to release the first real-time AI voice assistant model and a pretty slick demo (Moshi). Chris & Daniel discuss what this more open approach to a voice assistant might catalyze. They also discuss recent changes to Gartner’s ranking of GenAI on their hype cycle.
Leave us a comment (https://changelog.com/practicalai/278/discuss)
Changelog++ (https://changelog.com/++) members save 5 minutes on this episode because they made the ads disappear. Join today!
Sponsors:
• Plumb (https://useplumb.com/) – Low-code AI pipeline builder that helps you build complex AI pipelines fast. Easily create AI pipelines using their node-based editor. Iterate and deploy faster and more reliably than coding by hand, without sacrificing control.
• Motific (https://www.motific.ai/) – Accelerate your GenAI adoption journey. Rapidly deliver trustworthy GenAI assistants. Learn more at motific.ai (https://www.motific.ai/)
Featuring:
• Chris Benson – Twitter (https://twitter.com/chrisbenson) , GitHub (https://github.com/chrisbenson) , LinkedIn (https://www.linkedin.com/in/chrisbenson) , Website (https://chrisbenson.com)
• Daniel Whitenack – Twitter (https://twitter.com/dwhitena) , GitHub (https://github.com/dwhitena) , Website (https://www.datadan.io/)
Show Notes:
• Kyutai (https://kyutai.org/)
• Kyutai keynote video (https://www.youtube.com/live/hm2IJSKcYvo)
• Gartner Hype Cycle for AI (https://www.gartner.com/en/documents/5505695)
Something missing or broken? PRs welcome! (https://github.com/thechangelog/show-notes/blob/master/practicalai/practical-ai-278.md) | 775 | 11 | 0 | [Music], welcome to practical AI if you work in, artificial intelligence aspire to or are, curious how AI related Tech is changing, the world this is the show for you thank, you to our partners at fly.io the home, of, changelog.md 30 plus regions on six, continents so you can launch your app, near your users learn more at, [Music], fly.io welcome to another fully, connected episode of the Practical AI, podcast in these fully connected, episodes we try to connect you with, various things happening in the AI space, and connect you with maybe some learning, resources or talk about some subjects, that will level up your machine learning, game my name is Daniel whack I am, founder and CEO at prediction guard, where we're enabling AI accuracy at, scale and I'm joined as always by Chris, Benson who is a principal AI research, engineer at locked Martin how you doing, Chris doing great today it's uh dog days, of summer here in the US and uh it is it, is really hot and humid yeah super humid, and nasty I'm looking forward to AI, Control you know like weather control, from AI, and it will keep all of us at just the, right temperature right I can't see, anything possibly going wrong with that, of course not only only positives there, and that is in regardless that is in the, distant distant future of 2025 I'm sure, yeah exactly exactly let's let's focus, on the next two weeks for now uh which, is important I think one of the things, that caught me off guard this last few, weeks which you and I try to stay, plugged into various things and you know, maybe people think and listen to this, podcast that we're you know keeping, plugged in with every single thing, happening in the AI space but I was a, little bit surprised when I saw the, release um I guess I just hadn't really, been following along with what the, company or research lab qai was doing so, um this is a open research lab that, researches Ai and um they have funding, and some support in terms of, infrastructure and all of that but, they're a nonprofit research lab in my, understanding and um they actually so we, talked on a previous show we we kind of, got fooled a little bit or or maybe it, was you know a little bit of a fumbling, in terms of of marketing but it seemed, like when open AI GPT 40 came out you, know people were hyped because a lot of, the demos were voice-based but at least, you know at the time of that recording, I'm not sure all of what everyone has, access to in the the paid and unpaid and, Enterprise version but the actual voice, assistant for openai was not out and at, least as far as the release date of Q, Tai's Voice Assistant which is called, Moshi they were the first to actually, release a version of their voice, assistant which it's similar to in my, understanding what GPT 40 is is on the, multimodal side and that it is a, multimodal model so it's a realtime, multimodal model that supports a a voice, assistant and this research lab I think, it's like eight people or something like, that of course they have resources that, are supporting them right like this I, think it was a thousand gpus or, something they have resources obviously, but they were able to beat you know what, is now the the Goliath of the AI, space beat them to Market with this, realtime Voice Assistant which I think, took a lot of people maybe by surprise, or maybe some people were following it, closer and expecting it but I think um, in this sort of six Monon or whatever, time period it was when they were, working to get this out and beat the the, kind of Goliath of what is open AI which, I think in and of itself is is pretty, interesting it is I mean you know so, many uh try uh and some of the other, goliaths you know the the second tier, Goliath if you will are continually, trying to compete and they they may, touch it they may fall short I always, love hearing when a smaller group, especially if they're focusing on open, Solutions comes out and uh and is able, to do well so and they got a cool name, by the way yeah yeah and it is, interesting because this does run so, when you see the demo and and we can, pull it up here in a second and maybe, ask a few questions but when you see the, demo or the Prototype it obviously still, has some rough edges so I think you you, have some rough edges that aren't fully, kind of productized version like maybe, what is what you get with the open AI, Voice Assistant um in the forms that, it's in but it is very impressive also, because this is a model that I believe, it's models that are of a size that you, or I could run them on a even a single, GPU and they're going to open source, these models I don't I don't know what, the time frame is on that what exactly, that will look like what licensing all, of those things they do have a few talks, online so if any of the listeners know, that information and and I just haven't, run across it then then they can maybe, update us but yeah they they will be, open sourcing this which I think will, drive a lot more, experimentation and of course as we saw, with the first open llms that were, released with llama and other things, there was of course a huge explosion of, of innovation and and experimentation, going along with the release of the open, versions of those things and so I expect, that there'll be a similar thing with, you know these models and what I assume, will be other versions or other families, of these types of models moving forward, yeah I noticed you know going back to, your point about being able to to run it, locally potentially on a single GPU they, talk about in their press release uh, they just say compact uh Moshi can also, be installed locally and therefore run, safely uh on an unconnected device to, extend that a little bit I think that, you know there there are a lot of larger, organizations that are worried about IP, concerns these are topics that we've, covered quite a bit on the show in days, past so Moshi may very well find a home, in uh in corporate environments first of, all where they they don't want to send, information out and they want to get the, the advantage of that uh because it can, probably be run on a single GPU a lot of, edge devices make it possible so great, thinking there in terms of what's, possible and then finally uh thinking of, my own industry in the defense space um, since it can be run in an unconnected uh, or disconnected environment there's all, sorts of things from a from a uh a, government standpoint that they may be, willing toh to do so it's a great, strategy I I love I love hearing these, small companies that might be able to to, have a big impact uh in indust by, accommodating those those concerns well, Chris I find one piece of this whole qai, Moshi thing very interesting which is, almost like it feels a little bit like, deja vu because we back in whenever it, was I forget you know what what year, open AI came about is like there's these, big players in the AI space and they, were doing you know certain pre-trained, models and all of this stuff and robotic, things and and all of that and then open, AI came along and said oh we need a open, transparent nonprofit driven research, lab to really promote Innovation going, forward and of course as we have moved, forward through that we've seen open AI, kind of get away from that sort of pure, nonprofit status with the a little bit, more of a complicated corporate, structure right um which we've talked, about on on different shows but then, also you know just their release of, their work and their research and their, models and their data and those sorts of, things of course has has become very not, open and they of course have their own, reasoning behind that which at least, publicly they would they would say is, related to Microsoft sort of yeah well, at least publicly they would say is, related to safety of the use of these, models um of course you know there's, various people that might guess certain, other motivations Microsoft, yeah but um but yeah I I do find this, whole thing sort of like deja vu I don't, I don't know if if you're having the, same feeling here you and I both have, have a long history uh in the more than, six years now that we've been doing the, podcast of supporting open engagement uh, from different organizations uh whether, they're corporate entities or nonprofits, or whatever uh and we've seen that from, others I mean famously Yan laon talks, about uh he works for meta you know, which is Facebook's parent and talks, about nonetheless uh having open models, and all that and so we're we tend to, shine a spotlight on those organizations, that do that uh wherever possible we, certainly went through that because, we've been doing about just after we, started the podcast uh which uh was back, in 2019 or 2018 I believe actually 2018, um and about a year later open AI closed, up so we actually covered that in the, early shows you know it is what it is uh, they' they've done that they remain an, amazing corporate leader in the space, but yeah they did close all up and uh, and we tend to turn more spotlights, toward others like this so I'm pretty, excited to see uh what qai is is doing, and is able to do going forward here and, I hope this uh I hope they're able to, viably play against that top tier uh, competition I think that would be, wonderful to have multiple yeah do you, think that there's any chance for this, sort of like open research in the AI, space or in the technology space to, survive as a sort of bull workk of open, transparent research and open source uh, within the pressures that come of course, when you release this sort of technology, and you're a leader in the space and, there are actual dollar signs and, corporate concerns and certainly like, Partnerships that are necessary so you, know partnering with companies to do, this work is almost a reality I think in, the space because uh we talked about, this a little bit with the Stanford AI, index where they found that you know the, bulk of AI research is still happening, from the industry so I don't know what, are your thoughts is do they stand a, chance at staying staying the course, with this or I think there's certainly a, chance at it and I would argue uh it's, the same argument I've made in previous, shows where we talked on similar topics, is that we're seeing is the AI industry, is has been maturing these years uh at, an incredibly rapid Pace but we're still, seeing many of the things occurring that, we saw when when the software world was, really maturing over several decades and, the place where open source has really, really worked are in common touch points, where all organizations or many, organizations need a common thing and, they might build something, differentiated on top of that you know, for their revenue uh to drive, profitability but there's so much that, is underneath that point of, differentiation that they and many other, organizations can get the benefit out of, a lot of uh effort a lot of work a lot, of times they'll pay have paid employees, do it so there's a point where working, together and doing open stuff uh makes, sense for business and it drives, profitability it may not be your single, point of differentiation but if it's, anything under that why not you know why, not share the costs and uh pull, expertise for the best possible, foundations and so what I'm hoping is, that we continue to see that play out in, the AI space uh we're seeing you know if, you look at hugging face we've already, talked about the fact that uh a couple, of months ago they announced that they, were hosting a million models those are, all open source really really impressive, and so I think that there is a good, chance that a vibrant uh open Community, around AI can and will continue and it, will have a lot of corporate players, involved in it so I'm very optimistic in, that way, [Music], hey friends this episode of practical AI, is brought to you by our new friends, over at Plum Plum is a low code AI, pipeline Builder that helps you to build, complex AI pipelines super fast you can, easily create AI pipelines using their, node-based editor iterate and deploy, faster and more reliably than coding by, hand without sacrificing control, deployment is easy pipelines are live, API endpoints eliminate the the need for, constant code redeployment and debugging, by deploying complex AI pipelines as API, endpoints team collaboration is easy too, plums declared of node-based editor, enables you to build quickly while, empowering non-technical roles to, iterate on what you've done without, breaking it you can build Advanced AI, features get structured output every, time transform data and leverage, validated Json schema to create reliable, highquality structured output so Plum is, built for Builders early stage product, teams are using Plum to go from idea to, validation in record time to get started, go to use plum.com that's Plum with a b, as in plumber to request access today, that's us, pb.com again use plum.com, [Music], [Music], alrighty so as we uh as we change gears, just a little bit I had noticed uh a, couple of interesting things so I I, spend a lot of time talking to different, folks in the kind of in the Fortune 500, Fortune 100 world you know I I work at a, big company but I have a lot of friends, and former colleagues at other companies, and we chitchat about these things so, something has really come up a in a, whole bunch of conversations lately for, me and I thought wow if I'm talking, about it this much with with different, friends of mine it probably is uh is a, good topic to talk about on the show and, that's it's an interesting observation, and that is for those of you who are, familiar with the organization Gardner, uh and that organization does a lot of, um prediction and kind of identifying, different Technologies and things where, businesses uh can use them effectively, and famously they put out the gardener, uh hype cycle and what that is is it is, a life cycle for Technologies and they, basically across all technologies that, they track uh which is many they put, them on this hype cycle and track where, they are in their life cycle and the, short version of what that is it's a it, has a a steep upward curve that looks, like an ocean wave sort of that plunges, down into a trough behind it and then it, kind of comes up uh without so much, steepness Midway to kind of a, sustainable plateau and so what they, would argue is that for any given, technology there is an innovation, trigger which is this you know rocketing, up on amount of hype associated with the, technology uh and that it gets to a peak, which they refer to as the peak of, inflated expectations where it's really, high everyone's talking about it but, maybe not a lot of productive work has, happened yet super cool uh you can, probably already recognize how AI might, fit into this how we've you know with, all the things we've talked about over, time but then those expectations have, not been met and people become, frustrated with the technology and it, plunges down into what they call the, trough of, disillusionment and and that's where, they kind of go wow I thought that thing, was so great but boy it really didn't, pan out we wasted a lot of money on it, and uh it's just not it's just not, really worked out well for us but then, calmer Minds come along and they say, well wait a minute this technology has, some really good uses we just need to be, a little bit more practical pragmatic, about it and uh and and not lose our, heads over the hype and that's called, the slope of Enlightenment uh and that, reaches a point that's called the, plateau of productivity where where, basically for the long term a technology, lives out the rest of its life cycle, being a productive technology but, without all the craziness in the early, hype days so now that I've introduced, everyone to that to that life cycle I, going back to the conversation that that, I've been having uh repeatedly with, multiple people that I had noticed that, so many organizations especially large, organizations are just plowing money, into generative AI with mixed results, some are getting some decent results uh, with then you know within the context of, of it being early days in the corporate, sense but I noticed that after peing and, holding a peak on the hype cycle for, quite a long time generative AI is now, beginning to plunge down into the trough, of disillusionment and what that would, imply according to Gardner is that, people are beginning to get a bit, frustrated and I would say that's, panning out because I've noticed many, articles and social media posts over the, last few weeks that people have been, kind of going this isn't going to lead, to generative AI this isn't quite as, good as we thought it's not magic it's, all the things that you see with people, being a bit frustrated with it and those, are increasing in the number that I've, seen so it got us talking um about what, does that mean in a corporate sense, especially when you have a technology, plunging down into the trough of, disillusion and not only that but it's a, technology that has received a lion, share of funding relative to other, technologies that go through the hype, cycle it's the coolest of the AI you, know over the last couple years the, coolest of the AI uh tools in the, toolbox and uh with corporations always, lagging they're now plowing money into, it and yet expectations are falling and, so not getting to the point of it will, obviously find that slope of, Enlightenment and that uh plateau, productivity eventually what does it, mean over the next few months as we're, looking at organizations that are still, plowing money into uh generative AI but, maybe not in the most productive sense, or not as productive as they could given, the dollar value that they're putting in, so I've asked a lot of people what they, think of this Daniel what what are your, impressions of that you know it's an, interesting place to be if you're in, Corporate America or corporate anywhere, these days I do think it's interesting, and I think that in some ways some of, these feelings are are healthy in, particular what I mean is I noticed, earlier on so maybe in 2023 or you know, last fall still talking to a lot of, people with a misconception that oh we, have somehow what's going to happen is, we're going to get access to a large, language model or we're going to get, access to a foundation model in our, company and somehow that kind of equates, to a solution to them like this this, will now be a thing that solves problems, and I think that of course is a bunch of, baloney because basically a model does, nothing it's it's how you implement it, how you integrate it how you use it that, actually makes it a solution and so I, don't know how else to describe that, other than people thinking that AI would, provide a different typee of solution, than other Technologies which are, softwares that people deploy within, their companies right and so some of, this I think is really healthy in that, people are realizing oh wait a minute, there's still a need to think about how, we integrate a call to a large language, model in the context of a larger, engineering project and actually there, is engineering around the edges of the, integration of AI in some ways different, than traditional software engineering, and in a lot of ways the same whether, that be hosting services or testing and, evaluating outputs or virging the way, that we call these models or other, things there's a lot of those best, practices that are still really valid, from the software world and so to me, it's not so much and maybe this is just, just because I I of course have a vested, interest in the technology because I'm, building with it every day but I think, it's not so much a disillusionment about, AI functionality in the context of what, people are building over the next year, but disillusionment around how that, integration happens whereas before it, was sort of this fuzzy thing that we're, going to bring AI in and somehow that's, going to like solve a bunch of problems, without really an understanding of how, you would actually see return around, that now people are saying well yes, we're going to bring in AI we're going, to bring in llms but that's going to, live still in a software stack that we, have Engineers developing and we're, going to develop that on some life cycle, and yeah there's still going to be if, anything maybe increased engineering, spend because there needs to be extra, engineering around these models and so, it is enabling efficiencies it is, enabling net new kind of features or net, new products but these are still, products driven by software that, requires engineering and so that, realization I think is a really healthy, one and so maybe that thing that has the, disillusionment wasn't really ever a, real thing that could have been gained I, guess I think that's a fantastic Insight, I think in a perfect world if we can, help people along kind of get through, their own trough of disillusionment very, quickly to climb back up onto the slope, of Enlightenment by following that, guidance is is essentially what I'm, getting at before diving into it I I, know that over time as I've talked to, people it reminded me of uh Amplified, beyond what I've heard before but of, previous technologies that were supposed, to solve everything you know blockchain, was going to solve the world if you if, you recall blockchain was amazing we, were going to have it everywhere it was, going to be everything and by you know, having since reach that plateau of, productivity at the end of the life, cycle blockchain has a fantastic place, in the technology world and a Vibrant, Community but it of course doesn't solve, all things and I think people need to, realize that uh the same with these, kinds of models is that they they can do, that so I know that one of the things, that I'm trying to get people to do is, to get through their own trough of, illusionment quickly and start, recognizing in a really productive sense, how to fit it in with larger systems, we've always talked about it's really, the Software System around these models, that makes it all work that makes the, value for the user and even extending, that if you're not in the cloud it's the, hardware if you're out on the edge it's, all about what do you have on the, hardware and how does it integrate and, how does it integrate with the systems, you already have in place and what, special value are you expecting, generative AI to bring to bear that you, haven't already been uh trying to design, and solve for and so I think as people, really stopped and they kind of got out, of their their New Year's Eve party, moment and uh they said okay I'm an, engineer I need to start being an, engineer again and thinking about it and, they thought well maybe it doesn't solve, everything like I thought but I can, identify some pretty cool things that it, would help on value and I'm hoping that, people will start focusing on that and, bring engineering to your point back to, bear on this and solving it but solving, it in that larger ecosystem that, includes the overall stack that you're, in the software uh and since we're, moving, ever more out onto the edge into all the, devices that we use out there Beyond, just our cell phones that were always uh, ever present that we can find some good, uses so maybe this is a chance for a bit, of a Resurgence yes of engineering but, also I like this triggers all sorts of, like data sciency things in my mind, because as a data scientist operating, like in the in that industry for however, long it was the thing right it it was, about sort of choosing the right sets of, data tools and models to come up with a, solution or at least that's how I think, a lot of people viewed it and that may, have been a gradient boosting machine, plus a SQL data base plus a some sort of, data Pipeline and you know connecting, that into infrastructure and in, eventually into products that get out, into the world now some people might, view data science differently and and, have different views because it's sort, of an ambiguous term in and of itself, but I see one interesting thing on the, hype cycle that you were mentioning, there's a shorter time period that they, talk about this like composite AI, reaching the plateau than quote, generative AI which is interesting to me, and that I actually had to look up this, term because I have no idea what that, term means there's actually a number of, terms on the Garner height cycle that I, have no idea what they mean which I, wonder where they come from and I'm, right there with you so neither one of, knows what compos AI is so I'm sure that, there are a few people out there that, are very familiar with it and are, snickering at us and we welcome your, education and feedback on such keep, going though yeah but I looked up the, term and this appears to just be like, almost a term describing data science, which is just like using different types, of AI or machine learning together to, solve issues or create Solutions which, is sort of just is descriptive of data, science and kind of what it was for many, years so I don't know that it'll be, called data science maybe it's called AI, engineering I don't know but I do think, that we'll see kind of a return to this, idea of composite Solutions and a, multifaceted way of looking at at doing, these things not just with Gen but that, plugged in as an option into the, solution mix I couldn't agree more, despite being an AI podcast I know you, and I are always a little bit eye rolly, when it comes to all the hyper around it, we uh we try for our listeners to cut, through the hype uh and talk about it so, yeah a return to engineering and taking, advantage of some of these capabilities, in a holistic system uh to that is, highly productive and gives your end, users what they need is the way to the, Future, [Music], [Music], hey friends out shift Cisco's incubation, engine merges Innovation with the art of, possible a Launchpad for transformative, emerging Tech out shift Blends startup, agility with corporate strength to, develop nextg Technologies from the, groundup in AI Quantum Technologies, Cloud native and more their newest AI, Innovation Motif addresses a critical, challenge in the rapidly advancing world, of gen AI Bridging the gap between, concept and deployment this model and, vendor agnostic solution supports the, entire geni Journey from assessment and, experimentation Motif accelerates, deployment from months to days while, safeguarding against gen AI security, trust compliance and cost risks all, while empowering business function and, it teams to rapidly configure and user, assistance Power by organizational data, Motif provides advanced customizable, policy controls to prevent unauthorized, access to sensitive data and helps, ensure compliance throughout the entire, process with deep visibility into, operational and business metrics motivic, enables you to track Roi optimize costs, and make informed Decisions by offering, a centralized view moic deters Shadow AI, usage and empowers teams to innovate, responsibly so move beyond the, traditional constraints of AI, implementation utilizing AI deployment, that is both responsible and is, revolutionary ensuring your projects are, not just quickly launched but built on a, foundation of trust and efficiency visit, motif. a that is M, tfic c. a, [Music], [Music], well Chris I have um maybe a related, question for you which it's not exactly, related to the hype cycle but I was I, was at my local co-working space for uh, for a fundraiser on Friday night and, shout out to Matchbox co-working if, anyone's listening but yeah so that was, fun but I got into a number of AI, related conversations as I usually do, and one of the things that one of the, guys I was talking to mentioned was you, know there's a lot of people talking, about how this sort of wave of, generative AI this wave of AI in what, people are referring to AI now is being, compared to kind of like the surge of, it's like the new internet right like, when the internet was brought about and, the type of change that that created and, his point was sort of well that it, definitely created a a kind of New, Market the this space that was and is, the web and it wasn't just about, creating efficiencies and his point was, it seems like most people are using AI, to create efficiencies in in the, Enterprise whether that's you know, helping reports or automate certain, functionalities that interns were doing, before or analyze a bunch of documents, summarize those answer questions get a, quick access to information and from, from his standpoint these are all kind, of efficiency gains and not necessarily, creating any sort of New Market that, would be comparable to the huge shift, that happened when the when the web came, came about so I I was curious on on your, take of that may maybe it's slightly, related to the hype cycle stuff but uh, yeah I I think so there are two, different qualitative things there are a, lot of common traits between them but, because I'm slightly on the older side I, was an adult when the internet became, you know not when it was invented that, was actually was invented the same year, that I was born or a year before but um, at the point where it hit the general, public in a in a slight way I was in, college and by the time it became the, thing I was well into the workplace and, so you know that qualitatively the, Advent of the internet brought about a, brand new ecosystem upon which people, could do all sorts of new things I would, say it was like putting up it's like if, you're in a classroom it was like, putting up a chalkboard on the wall that, people can then go draw they can draw, mathematical equations they can doodle, they can do whatever they want but it, gave them a new medium upon which to, communicate and do stuff and interact, together and so it was that Baseline AI, is a bit different AI is it has a, similar revolutionary quality obviously, but it's expanding on that connectivity, and saying how can we get you what you, need faster and more, intelligently you know with Aid along, the way and so it's it's apples and, oranges but they're they're both in the, same fruit bowl a little bit of a, strange analogy there, yeah it's interesting that you bring up, the element of sort of creativity and, communication so there's probably some, parallels in the sense that I would say, there are many people treating these, sort of AI models and what they're, building with it as as a very creative, new I don't know if you'd call it a new, canvas on which they're painting but, definitely they're they're trying things, that are new and interesting maybe that, haven't been done before and are very, generative and but some of those things, even if you think about something very, much on the creative side like the udio, type of thing that is the music, generator that we talked about a while, back I think you could make an argument, well that is an efficiency Builder, because you could make a bunch of music, really quick for your YouTube videos or, a bunch of music really quick for your, ads or or whatever you're running online, but I think also some people are using, it as a creative element in and of, itself and doing maybe new and different, things or mixing things in ways that, people hadn't done in the past and maybe, there's thing other examples that are, that are better in my mind where kind of, it's almost a both and type of situation, so I'm I'm always maybe a sucker for the, third option where it's not like, clear-cut on one side or the other side, but this third option of yes it is about, efficiency gains but I think there is an, element of net new things that will come, out of the the AI space with these, models that maybe we are hard to predict, right now like they would have been hard, to predict in the rise of the web right, it was probably hard to predict what an, Amazon would become agreed when people, were kind of goofing around and making, websites to do this or that maybe it, would have maybe it wouldn't have but, like the level at which that sort of, company has shaped culture at large not, even just like Commerce but culture at, large you know maybe we just don't know, yet is one way to put it I I have a, couple of thoughts on that uh first of, all there is creativity in these AI, models and some people argue against, that even today like they may see but, they'll say they don't invent wholly new, Notions and stuff they they take things, that are already out there and they they, combine them and stuff like that and, there may or may not be Merit to that, but what I can do is I can compare it to, myself and other humans that I know um, and I'm an extremely creative human but, I'm creative in I have strengths in, certain areas of creativity and big, weaknesses in others and I've spent a, lot of time trying to compare myself to, these tools that I'm using uh in that, way and so I am very good at creating, out of nothing a software system in my, head and understanding all the right, things to put in place to do it even if, it's a a fairly new way of doing things, and that's a strength that I have I'm, terrible at drawing a beautiful picture, or painting and getting that out even if, I can Envision it in my head I can't do, that and what it's made me realize, seeing these tools that I'm using that, is that are producing these these, capabilities that we're all using all of, us listening to this are using every day, these days is it's made me really, question the sanctity of creativity and, I think uh at the end of the day I'm a, big believer that everything is, mathematical whether you agree or, disagree with it that you know we're, we're a biology we're based on chemistry, which is based on physics which is based, on math and that kind of science stack, that I tend to think of us as whether, something is silicon and producing stuff, from its capability or or is biological, in nature I spend a lot of time going, how special is what we as humans create, so maybe we just kind of acknowledge, that we're bringing things to bear and, these new tools that we're all using, every day brings things to bear and we, can be more productive and capable by, combining our talents and doing stuff so, I don't tend to be in either Camp I, don't tend to be in the uh this is, amazing new uh imagination from, computers my God what's the world coming, to and I don't tend to be in the ba, humbug this is just more of the same, I've seen this before and there's, nothing uh Magic about it I'm a little, bit in the middle and maybe a little bit, more nuanced than that I it's a long way, around to an answer I apologize yeah I I, definitely understand that and I think, from my own worldview and and even my, faith perspective I would think of a, sort of different special way in which, humans uh exist but at the same time the, we have created a lot of creativity with, the tools that we create and, technologies that we create and I think, there is something beautiful about the, fact that we are acting out as creators, creating things that are creative in and, of themselves right and so we're we're, kind of acting out the I don't know how, philosophical we've got on this show um, up to this this point but um but yeah we, can afford a moment here at theend, exactly this is the end of the show yeah, I would say it's it's kind of a, beautiful thing that we as human beings, are creative and we create things that, in and of themselves could be conceived, to be creative also and we co-create, with those things I think that's that's, really cool and I think that's an, element of what we've done with, technology over time and so yeah I think, my perspective is maybe we just haven't, seen what we are to co-create with this, technology um moving into the future and, how that will shape culture I think, that's going to be a longer time period, than maybe the one or twoe Gartner hype, cycle time period uh that we actually, see yeah this is shaping culture because, people know about it now but I I think, there's like a deeper way like people, knew about the internet at the sort of, hype of the internet coming out but, really how the internet would shape, culture and shape you know things like, what social media and other things have, done took a long time to realize so yeah, I I think that we have to wait a little, bit for that from my perspective great, perspective you have there and I would, encourage our listeners you know we had, a we had a little bit of moment of, finishing the show up with kind of, sharing our views on this but I think, this is important because we're all, going to see an increasing amount of AI, capabilities coming into our lives for, forever going forward at this point our, children are grandchildren the world is, changing faster now than it ever has so, these are thoughts that I hope you're, having as well evaluating how you see, yourself in this world sharing a world, with these technologies that are, increasing and if you haven't already I, hope you will join our slack Community, um where you can engage Daniel and, myself directly uh and share some of, your thoughts on how all this might work, going forward with creativity and with, these other topics we're having because, we'd love to hear your thoughts uh and, for what it's worth we build these shows, off of a lot of those conversations that, happen in the slack Community where, people are showing interest uh so please, engage us there share your thoughts uh, including the philosophical ones don't, be shy and I'm looking forward to, hearing what uh what some of you out, there are thinking yourselves cool well, thanks for having the discussion Chris, and hope you can uh uh have a good week, as you enter into more uh more fun AI, work sounds good Daniel uh stay cool in, the in the hot summer weather since we, don't have that AI climate control quite, yet yeah I'll see you next week all all, [Music], right all right that is practical AI for, this week subscribe now if you haven't, already head to practical AI FM for all, the ways and join our free slack team, where you can hang out with Daniel Chris, and the entire change log Community sign, up today at practical ai. fm/ community, thanks again to our partners at fly.io, to our beat freaking residence, breakmaster cylinder and to you for, listening we appreciate you spending, time with us that's all for now we'll, talk to you again next time, [Music] |
UCZbtcRQHGpU_tt2yM4C7k0A | NRrH9-kRMwc | Changelog | youtube#video | Vectoring in on Pinecone | Daniel & Chris explore the advantages of vector databases with Roie Schwaber-Cohen of Pinecone. Roie starts with a very lucid explanation of why you need a vector database in your machine learning pipeline, and then goes on to discuss Pinecone’s vector database, designed to facilitate efficient storage, retrieval, and management of vector data.
Leave us a comment (https://changelog.com/practicalai/277/discuss)
Changelog++ (https://changelog.com/++) members save 3 minutes on this episode because they made the ads disappear. Join today!
Sponsors:
• Plumb (https://useplumb.com/) – Low-code AI pipeline builder that helps you build complex AI pipelines fast. Easily create AI pipelines using their node-based editor. Iterate and deploy faster and more reliably than coding by hand, without sacrificing control.
Featuring:
• Roie Schwaber-Cohen – Twitter (https://twitter.com/roieschwabco) , GitHub (https://github.com/rschwabco) , LinkedIn (https://www.linkedin.com/in/roiecohen)
• Chris Benson – Twitter (https://twitter.com/chrisbenson) , GitHub (https://github.com/chrisbenson) , LinkedIn (https://www.linkedin.com/in/chrisbenson) , Website (https://chrisbenson.com)
• Daniel Whitenack – Twitter (https://twitter.com/dwhitena) , GitHub (https://github.com/dwhitena) , Website (https://www.datadan.io/)
Show Notes:
• Pinecone (https://www.pinecone.io)
• Pinecone | Blog (https://www.pinecone.io/blog)
Something missing or broken? PRs welcome! (https://github.com/thechangelog/show-notes/blob/master/practicalai/practical-ai-277.md) | 352 | 4 | 0 | [Music], welcome to practical AI if you work in, artificial intelligence aspire to or are, curious how AI related Tech is changing, the world this is the show for you thank, you to our partners at fly.io the home, of, changelog.md 30 plus regions on six, continents so you can launch your app, near your users learn more at, [Music], fly.io welcome to another episode of, practical AI this is Daniel whack uh I, am the CEO and founder at prediction, guard where we're enabling AI accuracy, at scale and I am joined as always by my, co-host Chris Benson who is a principal, AI research engineer at locked Martin, how you doing Chris doing great today, Daniel how's it going we're I know we're, recording leading into a holiday weekend, here we are and uh so many exciting, things La last week I got the chance to, briefly attend the the AI engineer, Worlds Fair which uh is sort of prompted, in in certain ways by our friends over, at the Laton space podcast and that was, awesome to see and of course a big topic, there was all things having to do with, with Vector databases rag uh all sorts, of you know retrieval uh search sorts of, topics and uh to dig into a little bit, of that with us today uh we have Roe, schwaber Cohen who is a developer, Advocate at Pine Cone welcome hi guys, thanks for having me today really, excited to be on the show yeah well I I, mean we were talking a little bit before, the show pine pine cone is from my, perspective one of the ogs out out there, in terms of of coming to the vector, search semantic search embeddings type, of stuff not that that concept wasn't, there before pine cone but certainly, when I started hearing about Vector, search and retrieval and these sorts of, things pine cone was already a name that, people were saying so could you give us, a little bit of background on on Pine, Cone and kind of how it came about and, and what what it is position-wise in, terms of the AI stack so Pine con was, started about four years ago give or, take and our founder Ido Liberty um was, one of the uh people who were uh, instrumental and founding um Sage maker, over at Amazon and had a lot of, experience uh in his work at Yahoo and I, think that one of the fundamental kind, of insights that he had was that the, future of pulling Insight out of data, was going to be uh found not exclusively, but predominantly in our capability to, construct vectors out of that data and, that representation that was produced by, neural networks was uh very very useful, and was going to be useful moving, forward I think he had that Insight way, before tools like uh chat GPT became, popular um and so that really gave pine, cone a great Edge at being kind of the, first mover in this space and we've seen, that the repercussions of that ever, since you know with the rise of uh llms, um I think people very quickly came to, recognize the limitations that llms may, have and it was clear that there needed, to be a layer that sort of bridged the, gap between the semantic world and the, structured World um in a way that would, allow llms to rely on structured data, but also leverage their capabilities as, they are and that is one of the places, where Vector databases uh play a very, strong role you know Vector dat, databases are distinct from uh Vector, indices in the sense that they are, databases and not indices right so like, an index may have basically is is, limited by the the memory capacity right, that the machine that it's running on, allows it to have whereas Vector, databases behave in the way that, traditional databases uh behave and in, the way that they scale of course, there's a completely different set of, challenges um algorithmic challenges, that come with the the territory of, dealing with vectors and high dimension, vectors that don't exist in the world of, just simple you know text textual, indexing and and colum data and that's, where like the secret sauce of uh pine, cone lies right it's its ability to, handle Vector data at scale but maintain, you know the speed and uh uh uh, maintainability and resiliency of a, database as you kind of comparing Vector, databases to indices and and then kind, of bringing that compared to that one of, the things that I run across still a lot, or people you know Vector databases are, are really you know incredibly helpful, now but there's still a lot of people, out there who don't really understand, how they F in you know they don't really, get it versus the nosql versus, relational databases or fine-tuning yeah, and so and they they hear you say it, does vectors and stuff like that could, you take a moment since we since we have, you as an expert in this thing and kind, of like lay out the ground work a little, bit before we dive deeper into the, conversation about what's different, about a vector database that is storing, vectors versus storing the same vectors, in something else like why go that way, for somebody who just isn't quite hasn't, really ramped up on that yet so the the, basic premise is you want to use the, right tool for the job right and the, basic difference right between a, relational database a graph database and, a vector database or a document database, for that matter right is the type of, content that they are optimized to index, meaning a relational database is meant, to index a specific column and create an, index that would be easily traversable, right and in scale um it would be able, to Traverse it across different machines, right and do it effectively right uh, graph database does the same thing only, its world live is nodes and edges right, and it's supposed to be able to build a, optimized uh representation of the graph, such that it could do traversals on the, graph efficiently in Vector databases, Vector databases are meant to deal with, vectors which are essentially long, high-dimensional set of numbers meaning, like you can you can think of an array, with a lot of real numbers inside of, that array and you can think of this uh, collection of vectors as being points in, a high-dimensional space and the vector, database is building effective, representations to find similarities or, geometric similarities between those, vectors in high dimensional space and, that means that basically it would be, very effective at given a vector finding, a vector that is very close quote, unquote to that Vector in a very large, space right so to do that like you need, to use a very specific set of algorithms, that index the data in the first place, and then query that data to retrieve, that similar set of vectors to the query, Vector at a small amount of time and, also being able to update or make, modification to that high-dimensional, Vector space in a way that is not cost, prohibitive right or or time prohibitive, right and that's like the the Crux of, the difference between a vector database, and other types of databases just to, draw that out a little bit more so from, your perspective like what would be if, you were to kind of explain to someone, hey here I've got one piece of text and, I'm wanting to match to some, close piece of text in this Vector space, what might be advantageous about using, this vector-based search approach and, these embeddings um in terms of what, they mean and and what they represent, versus doing like a you know tfidf has, been around for a long time I can search, based on keywords I can do a full text, search there's there's lots of ways to, search text um you know that concept, isn't new but this Vector search is, seems to be powerful in a certain way, from your perspective how would you, describe that yeah I think that the, lynchpin here is the word embedding, right the vector um search capability, itself is a pretty straightforward, mathematical operation that in and of, itself doesn't necessarily have value, right it basically it's like other, mathematical operations it's a tool, right the question is like where does, the value come from and I would argue, that the value comes from the embeddings, and we'll talk about what exactly they, are we just point a flag and say, embeddings are represented as vectors, which is why the vector database is so, critical in this scenario but why are, embeddings uh helpful in the first place, right so embeddings come from a, different set right like a a very very, wide set of neural networks that have, been trained on textual data and they, create within them representations of, different terms different surface forms, sentences paragraphs Etc that map onto a, certain location in Vector space the, cool thing about embeddings is that it, just so happens and we can talk about, why it just so happens that terms that, have semantic similarity have a, closeness in Vector space and that means, that if I search for the word queen and, I have the word King embedded as well in, my Vector database and I also have the, word dog, right because the word King is more, semantically similar to the word Queen I, will get that as a result and not the, word dog right and that allows me to, basically leverage the quote unquote, understanding of the world that machine, learning models and specifically neural, networks have right large language, models have of the world right in a way, that I can't quite leverage from other, modalities like tfidf and and other you, know bm25 Etc that like look at a more, lexical kind of perspective on the world, right and so when we talk about you know, practical use cases right like rag comes, up very very frequently and the reason, for that is because we are in semantic, space a user interacts with the system, in semantic space so that means that, they ask the system a question in, natural language we can take that, natural language and basically again, quote unquote understand the users, intent right and map it again into our, dimensional Vector space and find, content that we've embedded that has, some similarity to that intent right and, so we're not looking for a exact lexical, match but we're actually able to take a, step back right and look at the more, ambiguous uh intention and meaning of, the query itself and match to it things, that are uh semantically similar if that, makes sense would it be fair to say that, you're essentially because the output of, those structure being embeddings and, those are vectors and therefore you're, essentially storing it and operating on, it in a closer representation to how, they naturally would be and so you're, not doing a bunch of translation just to, fit it into a storage medium and and to, operate on it therefore it's going to be, quite a bit faster since you're is that, fair is that a fair way of thinking, about it perhaps we're we're in a way, we're compressing the representation, into something very small in a sense, right so you can think of an image for, example right an image would that that, could be like a megabyte big right we, can get a representation that in terms, of like its actual size in terms of a, vector is qu is order of magnitudes, smaller right and we can use that, representation instead of using the, entire image right to do our search now, it just so happens that again when we're, doing uh embeddings for images right we, get like that same uh quality where, we're not looking at you know we're not, looking at an exact match or like pixel, matching uh pixel to pixel with with, images that we have we can actually look, at the semantic layer meaning what is, actually in that picture so if it's a, picture of a cat we would get as a, result other pictures of cat that we've, embedded and saved in the database right, and that will come out of the, representation itself of the embeddings, that were a result of like say like a, clip model that we use to embed our, image so I don't know if it necessarily, means that like it simplifies things in, a lot of ways it actually adds a lot of, more oomph to the representation right, so you can actually match on things that, you wouldn't necessarily expect right, and that that's kind of like what the, beauty of semantic search in that sense, right is that um users can write, something and then get back results that, don't even contain right anything, remotely similar uh in terms of the the, surface form to their query but, romantically right it would be, [Music], relevant hey friends this episode of, practical AI is brought to you by our, new friends over at Plum Plum is a low, code AI pipeline Builder that helps you, to build complex AI pipelines super fast, you can easily create AI pipelines using, their node-based editor iterate and, deploy faster and more reliably than, coding by hand without sacrificing, control deployment is easy pipelines are, live API endpoints eliminate the need, for constant code redeployment and, debugging by deploying complex AI, pipelines as API endpoints team, collaboration is easy too plums declared, of node-based editor enables you to, build quickly while empowering non, technical roles to iterate on what, you've done without breaking it you can, build Advanced AI featur get structured, output every time transform data and, leverage validated Json schema to create, reliable highquality structured output, so Plum is built for Builders early, stage product teams are using Plum to go, from idea to validation in record time, to get started go to use plum.com that's, Plum with a b as in plumber to request, access today that's, us, mb.com again use plum., [Music], well roey I uh I really appreciate uh, also the the statement about adding, oomph to to your representations I think, that would be some there's some type of, good uh t-shirt that that could be, derived out of that I I see yeah for, listeners who are just listening on on, audio uh Ro's wearing a shirt that says, love thy nearest neighbor which is, definitely applicable uh to today's, conversation well this is great so we've, kind of got a baseline in a sense from, your perspective what a vector database, is why it's useful in terms of of what, it represents in these embeddings and, allows you to search through um you, mentioned rag we we've talked a lot, about rag on the show over time but, maybe for listeners this is the first, episode that they've listened to what, would be the kind of 30 seconds uh or or, some type of quick uh sort of remember, rag is X from from Rory right so I I, love quoting Andre kathi um with his uh, observation on um llms and, hallucinations so people when usually, people talk about rag they say oh Rags, sometimes hallucina and that's really, bad right and Andre kathi says actually, no they always hallucinate right that, they do nothing but hallucinate right, and that's really true right because, llms don't have any kind of tethering to, real knowledge in a way that we can, trust right uh we don't have a way to, say hey I can prove to you that what the, llm said is correct or incorrect based, on the llm itself right we need to go, out and look and search right and rag to, me is that opportunity where we can take, the user's intent we can tie it using a, for for example right a semantic, similarity search to structured data, that we can point to and say this is the, data that is actually trusted and then, feed that back to the llm to produce a, more reliable and truthful answer now, that's not to say that rag is going to, solve all of your problems but it's, definitely going to give you at least a, handle on what's real and what's not, what's trusted and what's not and it and, where the data is coming from where, those responses are coming from and it, shifts the role of the LM from being, your uh source of Truth to basically, being um a thin natural language rapper, that takes the response and makes it, palatable and easy to consume to a human, being great yeah I think a lot of people, have done a sort of maybe they've done, even their own demo with sort of a naive, rag um maybe pulling in a a chunk from a, document that they've loaded into some, Vector database they inject it into a, prompt and they get some useful output, one of the things that I think we, haven't really talked about a lot on, this show we we've talked about Advanced, rag methods to one degree or another but, I know pine cone um along with with, other Vector database providers you know, offer more than a simple just like, search that's the only F function you, can do there's a lot more to it that can, make things useful in particular like, having uh you know you mentioned pine, cone mentions kind of name spaces that, can be used metadata filters sort of, hybridized ways of doing these searches, could you kind of help our listeners, understand a little bit so they may, understand here's my user statement I, can search that against the database and, get maybe a matched document but for an, actual application like an application, in my company that I'm building on top, of this what are some of these other key, pieces of functionality that may be, needed for for a Enterprise application, or for a production application that go, beyond just the sort of naive search, functionality in a vector database yeah, for sure so we can take this one by one, so metadata is definitely like one of, those capabilities that uh Vector, databases have that are above and beyond, what a vector index would provide you, right and basically what they are is the, again the ability to perform a filtering, operation after your vector search is, completed and so you could basically, limit the result set to things that are, applicable in the application context, right so you can imagine different, controls and and selection Boxes Etc, that come from the application that are, more uh set in stone so to speak they're, not just like natural language they're, categorical data for example um and you, can use those to limit the the result, set right so that you hit only only what, you want that is something that is uh, very common to see in a lot of different, production scenario, and could you give maybe an example of, that like uh in a particular use case, that kind of you've run across like what, what might be those categories or what, just to give people something Concrete, in their mind yeah for example like you, can imagine a case where I'm not going, to name the the customer but like you, can imagine that the case where uh you, want to perform a rag operation but you, want to do it on a corpus of documents, but not on the entire Corpus but rather, on a particular project within that, Corpus so imagine that you have multiple, projects that your product is handling, um like finance and uh you know HR and, whatever uh engineering right and you, want to perform that search and then, limit it only to a particular project, and in that case right you would use the, categorical data that is associated with, the vectors that you've embedded and, saved in Pine Cone to only get the the, data for that particular project right, that is like a kind of super simple, example um but it can go beyond that, right and move into like the logic of, your application so like you can imagine, a case where you know you're looking at, um a movie a movie uh data set right, like and you want to search through, different uh plot lines uh of movies but, you want to limit the results only to a, particular genre right that's another, case right like we could just use, leverage U metadata you can think of um, wanting to limit the results that to a, time span right a start and end date, right things of that sort that kind of, like have to do more with the nature of, when and and how in what category the, vector belongs into and not specifically, the contents of the vector right so, that's one thing name spaces are another, feature that we've seen as being like, incredibly important for multi-tenant, kind of situation and multi-tenant rag, has become kind of like a very strong, use case for us and that's where you, know you you see a customer and that, customer has customers of their own and, not one or two but many many many and in, that case you definitely don't want want, to have all of the documents that these, that all of the sub customers uh have to, be collocated in one index and in that, case you basically uh break them apart, right so they're still in one index so, management of the index overall is, maintained Under One Roof but the actual, content and the the vectors themselves, are separated out physically from one, another in namespaces they're sort of, sub indexes to that super index and, that's another feature that we've seen, um as being super important to our uh, Enterprise customers as you're looking, at these Enterprise customers and with, maybe most Enterprises uh you know, getting into rag at this point at some, level and trying to find use cases for, their business to do that I know you, know my company and lots of other, companies are doing this what are some, of the ways that they should be thinking, about these different use cases when, we're talking about rag um and semantic, search and multimodal things that that, pine cone does what are good entry, Pathways for them to be thinking about, how to do this because you know they may, have come up with their own their kind, of own internal platform it might have, some open source it might have some, products already in in play but maybe, they don't have a vector database in, play yet and so you know how do they, think about where they're at when you, guys are talking to them and you're, saying let me you know we've C we've, been talking in the show so far about, kind of the value of the vector database, and the kind these use cases but not, necessarily kind of the an easy pathway, so how do you on board uh Enterprise, people to take advantage of the goodness, on this yeah that's an excellent, question and in fact it's like a quite, quite a big of a challenge because it, ends up being you know a straightforward, pipelining challenge that has existed, from the beginning of you know the Big, Data era right like is how do I how do I, leverage all the Insight that is locked, in my data in a beneficial way right and, the sad part about this story is that it, always depends on the specific use case, and it's hard to give a silver bullet a, sort of light at the end of the tunnel, is that we've recently published a tool, called the rag planner and its purpose, is to basically help you figure out what, do you need to do to get from where you, are to an actual rag application and, follow through all of the different, steps that are required in between right, and sort of like understand like from an, understanding of like where your data is, stored how frequently it updates like, what the scale of your data is it's etc, etc to the point where it could give you, some recommendation as to like what are, like the steps that you have to do like, in terms of do you build a batch, pipeline do you build a streaming, pipeline pipeline what tools should you, be using to do those things what kind of, data cleaning are you going to need to, do what uh embedding models are you, going to want to use to do this right, like how are you going to evaluate the, result of your R pipeline so all all of, these questions are pretty complex so, what I would say is as a general rule of, thumb first of all like you have to, evaluate whether or not rag is for you, right so for example there are a lot of, situations where you know rag may be the, wrong choice right because the data that, you have right and the actual capability, of answering end user questions based on, that data does not match up right and, that's how you get to see you know cases, where you know chat Bots sort of spit, out uh results that may seem ridiculous, but nobody catches it um and companies, get into a lot of hot water water, because of it right there are a lot of, scenarios where it's much easier to, start that journey and to sort of, develop the muscle memory that's, required in order to set these things up, in a lot of these use cases you see like, a lot more internal processes definitely, in bigger companies right where like, there's a a very big team that just, needs access to its internal knowledge, base um in an efficient way but um it's, not a system that is going to be Mission, critical right in any way so like if if, a person gets a wrong answer it's not, going to be the end of the world, nobody's going to get sued right and so, what I would say is there's definitely a, learning curve here for big, organizations for sure um it's usually, recommended to develop again that that, internal knowledge of what the, expectation versus the realities on the, ground is going to be to have like a, really good idea of how you assess risk, in those situations and most importantly, how to evaluate the results that are, produced by those systems right because, a lot of people are like okay you build, the rag system great and now produces, answers I'm done right like we're we're, everybody's happy that's farthest from, the truth that you could possibly be, right like these systems need to be, continuously monitored and feedback, needs to be continuously collected to, the point where you can understand right, like how changes in your data and the, way that you're interacting with it, changes in large language models that, you're implying are actually affecting, the end result right um are going to be, and how your users are actually, interacting with the system overall, right how all of these things kind of, coexist and happen together and are they, working in the way that you want them to, and of course you want to do that you, know in a quantitative and not, qualitative way right so like there's a, lot of instrumentation that has to go, into it I'm curious is a is a little, follow-up to that and obviously leaving, specific customers out of it are you, tending to see more internal use cases, of rag deployment to internal you know, groups of employees and stuff maybe from, a risk reduction are you seeing more of, an external I'm going to get this right, out to my customer and try to beat my, competition to it like where do you, think the balance is as of today I think, that there's a widespread and I think, that it's a journey right like I think, that like the more Tech native companies, that we see that are more I would say, forward-looking or you know, technologically uh adapt to kind of do, these things quickly are more ready to, not only take risks but take educated, risks in this space with the evaluation, that comes with it right so like these, are not just like let's set forget but, they actually know what they're doing um, in those cases you see them going out to, production with very big uh deployments, that is our bread and butter I would say, at the moment right with uh companies, that are more traditional um that have, been like that are not necessarily, getting Tech native you see a more uh, cautious sort of progression which is, only to be expected right like I think, that's kind of like natural to see well, Roe I have uh something that I I saw on, your website which it was new to my, knowledge which I think is also really, interesting one one of the things that, I've really liked in experimenting with, Vector database rag type of systems as, an AI developer is having the ability to, run something without a lot of compute, infrastructure maybe in an embedded way, or an Onis index something that I can, spin up quickly something that I'm don't, have to deploy a kubernetes cluster or, something to to uh or or set up a bunch, of kind a client server architecture to, to set up and test out maybe a prototype, that I'm doing and I and I see pine cone, is talking about pine cone serverless, now which is really really intriguing to, me just based on my experience in in, working with people um these sort of, serverless sort of implementations of, this Vector search I think can be really, powerful so could you tell us a little, bit about that and how that kind of, evolved and and what it is what's the, current state and and uh how pine cone, thinks about the serverless side of this, so serverless came about after we've, realized that tying uh compute and uh, storage together is going to um limit, the growth factor that our bigger, customers are are expecting to see and, it basically makes uh growth kind of, prohibitive in the space right and so we, had to find a way to break apart these, two considerations while maintaining uh, you know the performance characteristics, that our customers are are expecting and, are used to having from our previous um, architecture so essentially like, serverless has been a pretty big, undertaking on our side to ensure that, you know the quality of the database is, maintained but at the same time we can, reduce cost dramatically for customers, to just give you like an idea where like, um for the same cost of uh storing about, I don't know around 500,000, vectors before um you can now store 10, million right and that's a a humongous, difference right like it's an order of, magnitude difference I think that like, did accomplish that right like there was, like a lot of very clever engineering, that had to happen because again now, having compute and storage separated, apart means that storage can become very, cheap but on the other hand it requires, you to handle the storage strategy and, retrieval in a lot cleverer way we have, a lot of content on the website that, kind of delves deeper into how exactly, technically that was achieved and we, won't be able to cover that given the, time that we have but like the basic, premise is that you can now grow uh your, vector's index to theoretically Infinity, but practically to tens of billions and, hundreds of billions of vectors without, the cost of the expense becoming uh, prohibitive um which is the main drive, for us with our bigger customers and, also with smaller customers like um you, can start experimenting we have like an, incredibly generous free tier that, allows you to start you know like you, said right like if I'm just a developer, on my own testing things and trying to, understand how Vector database Works in, my world it's very unlikely that I'll be, able to tap the entire free toer plan, even several months in with many many, vectors uh stored right um and it will, work the same way that our Pro, serverless tiers work in terms of its, performance so it's not like a reduced, capacity performance in any way um so, you get get to feel exactly what it, would feel like and the effort that's, required to stand it up is minimal to, negligible right you just set up an, account and the SDK is super super easy, to use yeah and and am I understand sort, of representing things right like in, terms of the you know massive so there's, a massive engineering effort I'm sure as, you mentioned to achieve this because, it's not a trivial thing but in terms of, the user perspective like if people use, pine cone before and they're using pine, cone now you already mentioned the, performance is the is the interaction, similar it's just this sort of scaling, and sort of from the user perspective, scaling and pricing and maybe also you, could touch on so pine cone is people, might be searching for different options, out there and some of them would require, you to have your own infrastructure or, um some of them are hosted Solutions, pine cone at least in in its kind of, most uh typical form would be hosted by, you and yeah could you just talk a, little bit about the user experience pre, poost serverless and then also kind of, the infrastructure side like what do, people need to know and what are the, options around that in terms of what, happened pre and post so before um, serverless there was like a lot of, possible configuration choice that you, could do right like so like there was in, fact a lot of confusion with our users, you know like what exactly is the best, uh configuration for me should I use, like this performance kind of uh, configuration should I use like the, throughput optimized configuration what, exactly am I supposed to use and like, the pricing mechanism was a little bit, convoluted and I think that like serous, the attempt there was to simplify as, much as possible uh and to make it, really really dead simple right for, people to start and use but also grow, with uh with us right so again like I, said like the bottom line is you know, the external view into what pine cone, offers may have looked pretty similar, right so like if you're just a user you, may say like hey like I got like a, cheaper pine cone bill this month and, you know like I can store a lot more uh, always a good thing right always a good, thing right but not super amazing right, like but the end result is the question, is like what happens when you know you, can actually store a lot more vectors, right what what does that unlock for you, and and again I think that like the end, of the day right like the way that that, we see pine cone and this this may help, us uh kind of talk about like what's, next for pine cone is a place where your, knowledge lives right and and allows you, to build knowledgeable AI applications, right and having more knowledge is, always net positive right in that, context right so the assumption is that, like you know as AI applications grow, they accumulate more and more and more, knowledge and they become that more, powerful with any additional knowledge, that you can stuff into them and so, there's like actual value beyond the, fact that you can store more right like, and and it's cool right your application, actually becomes more powerful because, it can handle more types of of use cases, it has a better ability to be more, accurate and respond truthfully to a, user when they are interacting with it, and so I think that like in general like, there's like this uh blatant kind of, value that is only going to be apparent, once people really experience what it, means to have you know a million, documents uh that are stored in Pine, Cone versus 10 million documents that, are storn in Pine Cone and that effect, is going to be very powerful I think, that's the M majority of the of the, benefit that I see maybe that gets to, the next thing which I was G to ask, about which I also see the announcement, around pine cone assistance and I'd love, to hear more about that like what is, from of course um sometimes maybe that, can be loaded language also for people, in the in the AI space but in terms of, of this assistance functionality from, Pine Cone what are you trying to enable, and and where do you see it headed so, that has to do with the question that, Chris had before which is like what is, the journey right for customers right, and I think that like as a general, purpose that we had around assistant was, to reduce the friction between me having, a bunch of documents that I want to, interact with with an M or an AI in some, form and capacity to the point where, that actually works right there are a, bunch of ways of going about it right I, think pine cone wants to bring you know, on top of our very robust Vector, database a very smooth experience that, lets users uh really do very little and, get get all the value out of pine cone, without having to think too much about, it so for that purpose we don't only, have the ability to take your document, and then aded them and uh you know do, the endtoend process of creating that, completion endpoint for you right we're, also the ones providing the actual, inference uh layers as well right and so, it's not going to be again but if if you, asked like me like this questions like, how do you build a rag pipeline right, like a year ago even right I'd have to, tell you hey you have to go to like some, embedding provider you have to find, someone who would do like your uh you, know PDF F extraction or you know take, the data and chunk it and do all this, stuff right no more right like the, reality here is you can take a set of, documents throw them at this knowledge, assistant and the rest is kind of quote, unquote magic right it just happens for, you behind the scenes while maintaining, the quality that you want to get and at, the scale that pine cone can deliver, right which is again another, differentiator so like I said before, Pine con is built to withstand hundreds, of billions of documents right that of, vectors that you would store with us um, and still be able to produce responses, in a in a reasonable amount of time and, that's true for a knowledge assistant, because uh assistant sits on top of the, vector database so it sounds like that, may be uh a really good way especially, for small organizations you know we, talked about Enterprise and they have a, certain infrastructure and teams to go, with that but there's so many more small, organizations out there that have very, little in terms of of train people, necessarily uh to do that and they have, you know they don't have all the, infrastructure in place and they're, looking you know with assistance and, server list they're looking for simple, ways to on board and and get utility out, of it uh would you say that the, combination of serverless and assistance, and then maybe whatever they might have, in AWS or whatever platform that they're, using is kind of just made to gel easily, for them so they can get to something, working pretty quick yeah I mean at the, end of the day like if you think about, it right like like the process shouldn't, be as complicated as it is right it's, just that there are many parts to it and, nobody picked up the gauntlet of saying, like hey we'll just do it all you know, what I mean uh because all of it is, quite complicated to do right right and, so um yeah like I think that like, initially we'll see smaller, organizations kind of you know picking, that up because they don't have the, resources But as time moves along you, know you're going to have to ask, yourself even as a bigger organization, do I want to own this pipeline right is, that something that I need to own right, and what value am I getting from, actually owning all this right and so um, yeah like it would be interesting to see, so like this is a very very new product, still in public beta and it will be, interesting to see how the the market, kind of reacts to it and and and sort of, experiments with it but my bet is that, um as time progresses and uh knowledge, assistants themselves become more, capable doing things maybe Beyond rag or, Beyond Simple rag quote unquote um that, you know like more more and more uh, sophisticated organizations might want, to actually give it a try and that, really brings us maybe to a good way, that that we like to to end episodes, which is asking our guests to sort of, Look Look Into the Future a little bit, and not necessarily predict it because, that's always hard but to look into the, future and and kind of uh what what are, you excited about um it could be related, to Vector databases specifically or pine, cones specifically but maybe it's more, generally in terms of how the AI, industry is developing the sorts of, things that you're seeing customers do, that are encouraging whatever that is, what sort of keeps you um excited about, where things are headed going into the, the rest of this year I'm excited about, the fact that we're seeing sort of like, a Resurgence of what you would call, traditional AI kind of come back into, the fold um in the form of for example, graph rag I think that like the the, notion here is that you know for the, longest time and I think it's been since, like you know GPT, 35 um you basically saw like this like, uh I think over indexing on llms right, um for good reasons right like they're, super exciting they're they're very, powerful right and they can do really, really cool things right but um with, that said um it's as if every every, other technology that has ever existed, before just like dropped off the face of, the Earth and nobody nobody has ever, like talked about like okay wait so what, can we do with those things and llms, right like and where do llms fit in the, bigger picture I think that Vector, databases kind of like put llms in their, place a little bit in the sense that you, know what I mean like you're not, thinking of the llm as being the end all, be all like this is the only tool that, we need um I'm very excited to think of, llms as these operators or agents um, that can tap into the capabilities that, exist in other systems and I think that, what we're going to see more and more, and more is that people are going to, figure out like in what subset of the, ecosystem does each tool belong so what, set of problems that each tool solve um, for example like a vector database, solves like the problem of briding the, gap between the semantic world and the, structured World a graph database can, solve problems like reason like formal, reasoning over well structured data uh, relational databases can solve a whole, set of different problems that they used, to be solving like aggregation etc etc, and then uh you can imagine that llms, and agents can sit as sort of like an, orchestrating mechanism and a natural, language interface mechanism on top of, all those things together um and that's, what I'm excited to see like it's it's, kind of when like the the community as a, whole is going to like wake up from its, like llm fever dream and sort of realize, it like there's other things out there, um and and and realize that it has so, many more powers that that it could, yield um to make really exciting um, applications that's awesome well thanks, for painting that uh picture for us uh, Ro and and for taking time to dig into, so many um amazing insights about about, Vector databases and embeddings and and, knowledge management in general so uh, yeah appreciate what you all are doing, at Pine Cone and um hope to have you on, the show again to to update us on on all, those things thank you so much thanks, for having, [Music], me all right that is practical AI for, this week subscribe now if you haven't, already head to practical AI FM for all, the ways and join our free slack team, where you can hang out with Daniel Chris, and the entire change log Community sign, up today at practical ai. fm/ Community, thanks again to our partners at fly.io, to our beat freaking residents, breakmaster cylinder and to you for, listening we appreciate you spending, time with us that's all for now now, we'll talk to you again next time, [Music] |
UCZbtcRQHGpU_tt2yM4C7k0A | NRrH9-kRMwc | Changelog | youtube#video | Stanford's AI Index Report 2024 | "We’ve had representatives from Stanford’s Institute for Human-Centered Artificial Intelligence (...TRUNCATED) | 774 | 14 | 0 | "[Music], welcome to practical AI if you work in, artificial intelligence aspire to or are, curious (...TRUNCATED) |
UCZbtcRQHGpU_tt2yM4C7k0A | NRrH9-kRMwc | Changelog | youtube#video | Apple Intelligence & Advanced RAG | "Daniel & Chris engage in an impromptu discussion of the state of AI in the enterprise. Then they di(...TRUNCATED) | 548 | 5 | 0 | "[Music], welcome to practical AI if you work in, artificial intelligence aspire to or are, curious (...TRUNCATED) |
UCZbtcRQHGpU_tt2yM4C7k0A | NRrH9-kRMwc | Changelog | youtube#video | The perplexities of information retrieval | "Daniel & Chris sit down with Denis Yarats, Co-founder & CTO at Perplexity, to discuss Perplexity’(...TRUNCATED) | 369 | 1 | 0 | "[Music], welcome to practical AI if you work in, artificial intelligence aspire to or are, curious (...TRUNCATED) |
UCZbtcRQHGpU_tt2yM4C7k0A | NRrH9-kRMwc | Changelog | youtube#video | Using edge models to find sensitive data | "We’ve all heard about breaches of privacy and leaks of private health information (PHI). For heal(...TRUNCATED) | 292 | 1 | 1 | "[Music], welcome to practical AI if you work in, artificial intelligence aspire to or are, curious (...TRUNCATED) |
UCZbtcRQHGpU_tt2yM4C7k0A | NRrH9-kRMwc | Changelog | youtube#video | Rise of the AI PC & local LLMs | "We’ve seen a rise in interest recently and a number of major announcements related to local LLMs (...TRUNCATED) | 516 | 3 | 0 | "[Music], welcome to practical AI if you work in, artificial intelligence aspire to or are, curious (...TRUNCATED) |
UCZbtcRQHGpU_tt2yM4C7k0A | NRrH9-kRMwc | Changelog | youtube#video | New trend: RAG as a Service?! | "The Practical AI guys discuss a new trend in the AI world...\n\n#podcast #ai #machinelearning #data(...TRUNCATED) | 231 | 4 | 0 | "it kind of seems like people are, figuring out that generative models are, great at being kind of a(...TRUNCATED) |
End of preview. Expand in Data Studio
README.md exists but content is empty.
- Downloads last month
- 5