text stringlengths 0 2.46k |
|---|
[5699.34 --> 5704.80] So if you look at like the list of the hundred largest language models in the world, basically all of them are English or Chinese. |
[5705.46 --> 5705.54] Yeah. |
[5706.24 --> 5718.34] And, you know, so if you want to spread the benefits of this technology and the ability to kind of use and understand this technology to the world writ large, like not everyone speaks English and Chinese. |
[5718.34 --> 5721.80] And even the people who do often also speak other languages that they care about. |
[5722.42 --> 5726.94] So we're training, we've trained and released several Korean language models. |
[5727.56 --> 5735.62] We're currently training with the plan of releasing some Indic language models as well as some Romance language models. |
[5736.10 --> 5739.32] So, yeah, on the developing new model side, we do research like that. |
[5739.32 --> 5749.52] On the interpretability side, we do a lot of different stuff, understanding training dynamics, understanding how to evaluate language models, understanding how to kind of extract the best information from them. |
[5750.04 --> 5761.52] We recently started up some work on kind of red teaming them and trying to understand, you know, there's a lot of stuff out there right now about prompt hacking, about how people are trying to put filters on language models and they're kind of not really very successful. |
[5761.52 --> 5770.60] And trying to understand, like, what the dynamics of that is like, whether you can build meaningful safeguards around these things or whether it's always going to be subverted. |
[5770.86 --> 5772.26] We do a lot of work like that as well. |
[5772.94 --> 5773.32] Very cool. |
[5773.96 --> 5775.40] Well, thanks for coming on the show, Stella. |
[5775.62 --> 5776.68] Yeah, it's a pleasure. |
[5777.12 --> 5778.98] It was awesome having this deep dive with you. |
[5779.00 --> 5779.42] I love that. |
[5779.52 --> 5779.90] Thank you. |
[5780.18 --> 5780.84] Great to meet you guys. |
[5780.84 --> 5808.42] Yeah, so if you'd have told me a few years ago that I'd be going to an open source summit and talking about AI in open source at this level from Cody, a coding assistant to Databricks and training models on small data sets to Stella's work and the Luther AI's work on open AI research and all thes... |
[5808.42 --> 5816.74] That it'd be real, that it'd be touchable, that it'd be usable today to transform my work, to transform your work, to transform the world around me. |
[5817.12 --> 5819.64] I would not have believed it, but it's true. |
[5819.94 --> 5821.88] We're here and this show was awesome. |
[5822.02 --> 5823.14] So hope you enjoyed it. |
[5823.44 --> 5831.66] Once again, a big thank you to our friends at GitHub for sponsoring us to go to this conference as part of Maintainer Month. |
[5832.22 --> 5836.32] There is a small bonus for our plus plus subscribers. |
[5836.78 --> 5838.24] So stick around for that. |
[5838.24 --> 5840.60] If you're not a plus plus subscriber, it's too easy. |
[5841.10 --> 5843.34] Changelog.com slash plus plus. |
[5843.54 --> 5844.74] We drop the ads. |
[5844.92 --> 5847.00] We obviously give you bonus content. |
[5847.34 --> 5849.38] We bring you a little closer to the metal. |
[5849.56 --> 5852.10] And the best part, you directly support us. |
[5852.50 --> 5854.74] Ten bucks a month, a hundred bucks a year. |
[5855.26 --> 5857.50] Changelog.com slash plus plus. |
[5858.14 --> 5858.62] That's it. |
[5858.68 --> 5859.32] This show's done. |
[5859.54 --> 5860.42] Thanks for tuning in. |
[5860.42 --> 5862.48] We will see you on Friday. |
[5868.24 --> 5898.22] We'll see you on Friday. |
• Introduction to All Things Open 2023 and the podcast's sponsors, including Neon |
• Interview with Nakita Shamganov, co-founder and CEO of Neon, on the company's mission and technology |
• Discussion on the modern developer experience and how Neon aims to perfect it |
• Overview of Neon's features, including on-demand scalability, bottomless storage, and database branching |
• Conversation about the response to Neon, with onboarding of 2500 databases per day |
• Brief history of Postgres and its current popularity among developers |
• The Postgres community is aging and facing challenges in transitioning to new leadership. |
• The core contributors to Postgres are mostly men in their 50s and nearing the end of their careers. |
• A transition in leadership is necessary to ensure the project's future. |
• The introduction of multi-threaded architecture could be beneficial but would require significant changes to the software. |
• The existing code and ecosystem would need to be adapted to be thread-safe. |
• Libraries and software now often include thread-safe versions, making thread safety a less significant concern. |
• Difficulty in detecting thread safety issues in open-source projects |
• Governance and decision-making process in Postgres community |
• Challenge of introducing large changes to Postgres |
• Potential benefits of making the storage manager API more pluggable |
• Status of patches submitted by Neon to Postgres community |
• Concerns about Neon's competitive advantage being compromised by open-sourcing code |
• Postgres community's history of falling behind on features and being caught up by others |
• Discussion of Neon database's architecture and features |
• Separation of compute and storage, and its benefits |
• Use of Postgres as the underlying database |
• Integration of extensions such as pg_vector and PostGIS |
• Potential for geo-distributed Postgres deployment |
• Current limitations and future plans for geo-distributed Postgres |
• Discussion of branching and its unique capabilities |
• Overview of Neon's storage system and its role in the architecture |
• Mention of other exciting developments in the Postgres world, such as pg_vector and asynchronous IO |
• Personal interest in reviewing and integrating patches for asynchronous IO to improve Neon's performance. |
• Application monitoring platform that shows what's slowing down the line of code and makes performance monitoring actionable |
• New approach to performance monitoring that groups error codes and gives users everything they need to solve errors |
• Comparison to traditional performance monitoring, which can be time-consuming and requires a lot of context |
• Trial of new performance monitoring features, which involves setting up transaction information and configuring the SDK |
• Web Assembly discussion, with Robert Abukalil agreeing that it's a heavy-duty tool with limited practical use beyond specific needs |
• Robert's experience with Web Assembly in bioinformatics and his concerns about overhyping its potential uses |
• Web Assembly (WASM) capabilities and limitations |
• Using WASM for server-side applications, such as plugins and sandboxing |
• Bringing bioinformatics tools to the web |
• Using WASM to power interactive tutorials for command line tools |
• Bioinformatics definition and applications |
• Limitations of running heavy-duty analysis in the browser |
• Pronunciation of WASM (wasm or wasm) and its origins |
• Discussing a past argument on the show about "jiff" vs "gif" |
• Bioinformatics applications being moved from desktop to web |
• Types of applications suitable for web-based bioinformatics tools |
• WebAssembly's limitations and potential for performance improvements |
• Converting tools from existing languages to web-based versions |
• Optimizing webAssembly performance by reducing data exchange between JavaScript and webAssembly |
• CLI tutorials in the browser, and emulating a full Linux environment |
• Using an open-source project called v86 to emulate a CPU and boot a BIOS in the browser |
• Discussing the limitations of emulating complex systems, such as BIOS and hardware, in a browser environment |
• The "uncanny valley" of emulation, where the emulation is not perfect, leading to limitations and performance issues |
• Potential uses for emulating complex systems in a browser, such as tutorials and educational resources |
• Projects that use emulation in a browser, and the potential for tutorial sites to use this technology |
• The idea of emulating specific operating systems, such as Debian, to demonstrate installation processes and features |
• Using xterm.js to create an emulated terminal environment for interactive tutorials and exercises |
• Creating tutorials for general use, not just bioinformatics, to teach basic tools like awk, grep, and git |
• Creating tutorials and bringing text-based tutorials to life using interactive sandbox.bio |
• Embedding tutorials and demonstrations on a website |
• Using Web Assembly to run tools directly in the browser |
• Authoring own tutorials and embedding them on the website |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.