text stringlengths 0 2.68k |
|---|
**Salvatore Sanfilippo:** Okay, an example... Sometimes when I take weight, I want to lose this weight. Normally, I do calorie jacking with my \[unintelligible 01:07:27.17\] or other applications, and stuff like that. However, it's so boring. And I do that for 10 years at this point, but still it's boring. So I'm writi... |
So what the bot does is that based on my query, it asks Redis for similar items. Then I take the similar items of foods that are similar, and it takes that in the context of the LLM when asking the question. "The user reported eating this thing. Based on this table, don't invent the macronutrients. Please tell me what ... |
**Adam Stacoviak:** Key phrase there, or key word there, Jerod, was please. |
**Jerod Santo:** He did say please. |
**Adam Stacoviak:** He did say please. \[laughter\] The debate is still out. |
**Jerod Santo:** It's only one token in English, you know? I don't know about in Italian, but... It's only one token in English. Okay, that's cool. That's very cool. I'm sure there'll be countless other people with ideas on how they can leverage this to do cool stuff, whether it's commercially, or healthily in their ow... |
**Break**: \[01:09:24.24\] |
**Adam Stacoviak:** I don't know if this maps to this, but there's a particular problem I'm trying to solve for me. So as you may know, Salvatore, we are a sponsored podcast. And so we have brands, like Redis - which have never sponsored us - that we would happily partner and share with the world, the developer world, ... |
So I feel like what you're talking about with this vector embeddings is like "Okay, in that world, building my own GPT or my own thing like that is possible because you've got these embeddings in there", and whenever I prompt the LLM, it's going back to those embeddings and saying, "Okay, I know this costs Y, and we're... |
**Salvatore Sanfilippo:** Yeah, yeah. In general, if you want to retrieve anything -- for example, another example that... For your use cases particularly, one of those systems that have a very large context window, like Claude Sonnet, and that don't use RAG and see all the files in the context could be the best spot f... |
**Adam Stacoviak:** I want that. |
**Jerod Santo:** \[laughs\] "I want that so bad." |
**Adam Stacoviak:** The math being off is just like driving me crazy. I just did it today, I was like, it's accurate 95% of the time, and then one time it's like, it does terrible math. I'm like "I could never charge our customer that or our partner that, because that's not the real number." And so I've got to check it... |
**Salvatore Sanfilippo:** Another thing that you can do is to specify to the model that you want all the calculations performed using a program. This also is a trick that works. |
**Adam Stacoviak:** Okay. Interesting. |
**Salvatore Sanfilippo:** \[01:15:43.16\] And Claude unfortunately cannot run Python, but it writes JavaScript and executes it inside your window of your browser, and then takes the calculation back. But to make another example of vector sets, imagine that you want to do something like face ID. So basically, the iPhone... |
A few years ago there was one guy that did this in Hacker News. It downloaded all the data, all the comments of all the users, and created vectors with the most used 10,000 words, setting the number of occurrences for each word. So in a given vector, there were many zeros, but other things valorized. And using this and... |
**Jerod Santo:** Cool. What's your YouTube channel called? |
**Salvatore Sanfilippo:** Salvatore Sanfilippo. |
**Jerod Santo:** \[laughter\] Keep it simple. |
**Salvatore Sanfilippo:** The coding videos are in English, a lot of talking videos are in Italian, but what I do is to take the transcript, pass it to Claude Sonnet, and it cleans up the Italian, and then the auto-translate to English works very well. There are other people that don't talk Italian following me in the ... |
**Adam Stacoviak:** So it sounds like you're a Claude Sonnet fan over Open AI's models. Like, you're hanging out in there. |
**Salvatore Sanfilippo:** Definitely. |
**Adam Stacoviak:** Okay. And how do you get around this limitation of what we talked about before when you volley back and forth? I don't know, you have a long chat history. How do you get around that? You just restart it? That's your way? |
**Salvatore Sanfilippo:** Yeah, in general, I use AI a lot for coding, but really big time, but in a strange way. Most of the time I am in my terminal, writing my code myself. When I've got a problem, I go there and often my prompt is like "Don't write any code. Let's discuss about this", because I use it like an impro... |
**Adam Stacoviak:** \[01:20:26.17\] Gotcha. |
**Jerod Santo:** So are you using this outside of your editor, or inside of your editor? |
**Salvatore Sanfilippo:** Not integrated, because for me this is a way to avoid becoming lazy, basically. I want to be sure that 100% of the code that I use, even if some code is written by the LLM, I understand what's happening. But I use it a lot more to track bugs. Like, I have a bug, oftentimes Claude Sonnet, if I ... |
**Adam Stacoviak:** So good. Yeah. |
**Salvatore Sanfilippo:** Incredible. Like, five hours of debugging, five minutes. Because it will say -- after refactoring a few days ago, I forgot a statement; I removed the statement, but the statement removal didn't cause any error in the compilation, but it just created a memory error. And what they do in this cas... |
**Jerod Santo:** Well, I'm going to have to try Claude Sonnet now. I've never tried it. I've been happily doing the other things. I don't know if "happily" is the right word, but I've been doing other things, and that's one that I just happily ignored so far. But you've convinced me. I need to give this one a shot. May... |
**Adam Stacoviak:** I can toot it a little bit, the horn of Claude a little bit. So I saw somewhere - I think it was on YouTube, or something like that, someone was saying it's really great for building documents. So I needed to create an agreement... And I knew what I wanted to say, but the way Claude works is it will... |
**Salvatore Sanfilippo:** You see it removing the lines, like pressing the cancel key, because it uses the artifact as a tool. |
**Adam Stacoviak:** Yeah. |
**Salvatore Sanfilippo:** So it can remove, edit, change parts... And also, between Claude 3.5 and 3.7, the generation output length was increased like 10 times. So now it can generate like 1,000 lines of code in one shot or even more, which is pretty impressive. |
**Adam Stacoviak:** Wow. That is impressive. |
**Salvatore Sanfilippo:** It's a very great system. Also, when Dario Amodei started the company, they were also extremely focused on the ethical part, the safety part, which is also a plus... Even if I believe that what they released a couple of days ago, Claude Code, which is agentic coding, I believe this will start ... |
**Adam Stacoviak:** \[01:24:12.24\] Ignite, yeah. |
**Salvatore Sanfilippo:** A lot of firing or mis-hiring of programmers. Because it's -- |
**Jerod Santo:** It's that good? |
**Salvatore Sanfilippo:** Yeah, it's -- Claude Code is the first thing I see which is not conceived to empower the programmer that wants to do more, but to replace. |
**Jerod Santo:** Ahh...! |
**Adam Stacoviak:** So much for ethics... I'm just kidding. \[laughter\] |
**Salvatore Sanfilippo:** I understand that it's inevitable that sooner or later it's going to happen, but -- |
**Adam Stacoviak:** Someone's gonna do it, yeah. |
**Salvatore Sanfilippo:** ...what is my interest is that if we have to go forward like that, that governments in all the world understand that they have to find ways for the people to pay their bills. And then it's okay if machines want to code; as soon as people are safe, I am okay with it. |
**Jerod Santo:** Well, on that demure note, anything that we haven't talked about that doesn't include the career extinction of us software developers...? \[laughter\] Anything else you want to mention, Salvatore, before we let you go? |
**Salvatore Sanfilippo:** Just that -- I see that, in some way, even if we say that our field in some way in these 15 years changed. However, imagine something as big as AI in the hands of other industries, for example pharmaceutical industry or finance. They will never -- it will never be like that that we have most o... |
Also, if you think at the price of Open AI and Anthropic right now, $20 per month is very low. It's very democratizing this tool. So in some way, I continue to be optimistic. I am more about the \[unintelligible 01:26:40.27\] side, which I believe that in the short run the problem that we will face with AI - as a socie... |
**Jerod Santo:** Well, we're excited that you're back at Redis... \[laughter\] We're excited for the subtle changes, the incremental changes that you're making to help Redis be more useful to more people... Thank you once again for creating such an awesome system and for open sourcing it, so that the world can take adv... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.