วีดีโอ

Great work

great vid! I love the fact he still codes that much minor feedback: reduce the amount of frame changes, the content is already fun as it is (when you have a great guest), no need for the production to go wild to keep viewers's attention (i think, at least for me)

her tweet is code for infra team 😅😅

Great talk! Google really have great creative people working for them.

Steven Johnson's description of using DevonThink from Where Good Ideas Come From surely inspired a chunk of the design

Nice work & explanation.

thanks for having me!

Thank you so much!! I never understood "The Turnover Village" until NotebookLM explained it to me.

i love that guy! Pls have him again on the platform. The depth of his understanding from low level to ML application including the financial side, is very impessive.

MashAllha

Good stuff! 🤓 Found you on X.

Love you Dear

f love how NotebookLM podcast begin with Deep Dive opening 😆

Ha ha brilliant 😀

This is really helpful for me. Thank you.

🎉🎉😮

Lets not. This guy is approaching it to be lazy. To not have to do "unimportant" stuff.

Full writeup: www.latent.space/p/josephine-teo

Insight dense podcast, relistening and taking notes

Wow, talented dude! Great interview!

No echo on Dylan which sounded great but unfortunately the other audio was kinda echoing but amazing podcast in all other aspects 👏 👏 👏 👏 👏

seems like Drew is deep into music.

That looks like a Sunny Hills Pineapple cake

sorry we screwed up the video recording! see the full writeup as well! www.latent.space/p/braintrust

see also Amgad's writeup on Whisper amgadhasan.substack.com/p/demystifying-openais-new-whisper

🇧🇷🇧🇷🇧🇷🇧🇷👏🏻, I'm eagerly awaiting the release of real-time video for Plus users from OpenAI, as it was originally mentioned as part of the ChatGPT Omni update, which sadly never reached us. This feature will be revolutionary, enabling us to tackle a wide range of daily tasks more efficiently. Real-time video integration within ChatGPT would greatly enhance productivity by allowing for interactive, dynamic assistance and more streamlined workflows. It would be especially useful for tasks like desktop sharing-being able to visually assist and collaborate on real-time activities is just phenomenal. I hope this feature rolls out soon, as it could drastically improve how we approach everyday challenges.

Asking Claude: Read the provided transcript. Provide lists for each of the following that are mentioned: APIs Tools Techniques People Organizations Problem Domains ----------------------- Here are the lists based on the provided transcript: APIs: 1. Real-time API 2. WebSocket API 3. Chat Completions API 4. Assistants API 5. Function calling API 6. Structured output API Tools: 1. ChatGPT 2. GPT-4 3. GPT-3.5 4. O1 (OpenAI's new model) 5. Cursor (coding tool) 6. Genie (coding AI) 7. Cognition (AI tool) 8. Harvey (AI tool) 9. Case Text (AI tool) 10. Notebook LM (Google's tool) 11. Claude (Anthropic's AI) 12. Gemini (Google's AI) 13. Waymo (self-driving car technology) 14. Code Interpreter Techniques: 1. Fine-tuning 2. Distillation 3. Prompt caching 4. Vision fine-tuning 5. Model evaluation 6. Iterative deployment 7. Chain of thought reasoning 8. Function calling 9. Structured outputs 10. System prompts People: 1. Sam Altman 2. Kevin Weil 3. Roman Hewitt 4. Michelle Pokras 5. Simon Willison 6. Alistair Pullen 7. Olivier Godemont 8. Ilya Sutskever (implied) 9. George Hotz 10. Jon Frankle 11. Chris Lattner 12. Jeremy Howard Organizations: 1. OpenAI 2. Anthropic 3. Google 4. Microsoft 5. Cosine 6. Cognition 7. Harvey 8. Case Text 9. Twilio 10. LifeKit 11. Agora 12. Waymo 13. Twitter (former) 14. Coinbase (former) 15. Y Combinator Problem Domains: 1. Natural language processing 2. Computer vision 3. Speech recognition and synthesis 4. Code generation 5. Language translation 6. Autonomous systems 5. Artificial General Intelligence (AGI) 6. AI safety and alignment 7. AI ethics and responsible deployment 8. Developer tools and platforms 9. AI-assisted software engineering 10. Real-time AI interactions 11. Multimodal AI (text, voice, vision) 12. AI agents and automation 13. AI in government and public services 14. AI for scientific discovery 15. AI user interfaces and experiences

Interesting.

high quality video is now up here x.com/marksaroufim/status/1841277387834830876

Is the host the guy in Silicon Valley who played the supermarket employee who quit his job to build an app which could find a parked car?

thanks for recording and sharing

Simple answer is Yes, the problem with most of the major contributors is they're trying to build without knowing or understanding what the best end would be ( building thinking about the current without the end in mind). Feeding a machine all of the data makes it knowledgeable, but it will always lack true intelligence until it realizes that there are hundreds if not thousands of ways to solve a problem and it's able to narrow down the best course of action to get the best result.

AYYO OP really published an AI paper with the term reverse-harem in it.. *FISTBUMPS*

What's the link to the paper mentioned in this talk (timestamp: 23:45)? Great content!

Great pod. Thx!

harrison looks a bit sad

3D Variety analysis point of point the point.

3D Python Pylogik. Point of photo Python.

Cool to hear from a systems engineer, but agree with other commenters - she’s not working on AI or AGI ;)

13:00 + 20:15 -> I guess we're the avengers-pandas now :))

Intelligence creates knowledge and improves knowledge over time.

like and subscribe! www.latent.space/p/shunyu

So good! Love these reflective style presentations.

Great reviews BTW, Synthetic data can be llm generated or by traditional programming.

16:05 "I deleted all the streams, made everything single-threaded because we ended up getting all kinds of weird race conditions and errors and so on and I just didn't want to deal with it." Andrej,I hate to be that person, but Rewrite it in Rust! 😊

First. Great interview. BTW I get leaks from my exemplars..too influential. I ask Claude for his hidden reasoning trace and he helps. My experience, the best reasoning trace varies by problem because it follows that domain's problem solving process.

Thanks for recording this!

Wow!

Andrej blew my mind there at the end: "Python and PyTorch and everything else is just a crutch because we humans are finite with finite knowledge intelligence and attention. But actually, don't you want to write all code in CUSTOM CUDA kernels?" 🤯 CUDA MODE! Let's make the GPUs goo BRRRRRRRRR!

❤