Weeknotes 2023-04-03
,if my editor reads this: everything’s totally fine --Ethan Marcotte
Not much work, but a fair share of paperwork, and finally traveled to Erlangen. New django-cast-release and applied with django-cast to the prototype fund🤞. Downloaded the LLaMA weights, and hope to find some time to play around with LLMs locally next week.
Articles
- AI-enhanced development makes me more ambitious with my projects | Example is a proxy fetch function to submit past chatGPT conversations to datasette, cool!
- Cheating is All You Need | First Steve Yegge post I've seen in a long time, welcome back 😁
- Learn Images | I guess I have to
- Rust's Golden Rule
- My Favorite CSS Snippets
- Defaulting on single page applications (SPA)
- Pause Giant AI Experiments: An Open Letter | I think this is dumb, but some smart people did sign it? wtf...
- Unravelling
del - Illustrating Reinforcement Learning from Human Feedback (RLHF) | How to get from GPT3 to chatGPT
- Aligning language models to follow instructions | InstructGPT
- Learning from human preferences | RLHF explanation from openAI
- Think of language models like ChatGPT as a “calculator for words”
- The age of average | Hmm, for me it's more like regardless of how hard I try to zig, I still zag.
Weeknotes
- Weeklog for Week 12: March 20 to March 26 | Johannes
- Weeknote for Week 13 | Bithive
Mastodon / Twitter
- I wish elon had bought LinkedIn
- 1/The call for a 6 month moratorium on making AI progress beyond GPT-4 is a terrible idea.
- Last weekend's GPT-4 Hackathon at the @agihouse_ai had: ✅ 170 AI founders and researchers ✅ Talks by @eladgil , @ClementDelangue ( @huggingface ), @8enmann ( @AnthropicAI ), @robertnishihara ( @anyscalecompute ), & @BorisMPower ( @OpenAI ) ✅ 50 projects Projects that stood out👇
- llama.cpp just got access to the new Copilot for Pull Request technical preview by @github
- music is mostly just really loud math
- a big deal: @elonmusk , Y. Bengio, S. Russell, @tegmark , V. Kraknova, P. Maes, @Grady_Booch , @AndrewYang , @tristanharris & over 1,000 others, including me, have called for a temporary pause on training systems exceeding GPT-4 | Uhm, no.
- To run the released model with latest llama.cpp, use the "convert-unversioned-ggml-to-ggml" python script and apply the following patch to llama.cpp The latest llama.cpp offers significant performance and accuracy improvements in the inference computation
- Made some quick notes on how to use the OpenAI Python library to make ChatGPT API calls and stream out the response tokens as they arrive
- Announcing the Local LLaMA podcast 🎙️🦙 In today's episode we have LLaMA, GGaMA, SSaMA and RRaMA joining us to discuss the future of AI
- Self-attention is the primary building block of large language models (LLMs) and transformers in general. But, how exactly does it work? 🧵 [1/8]
Open Source Funding
- Zuschüsse: Direkte finanzielle Förderung | NRW
- Time to give back to the community | Open Source Fund
Software
- BlenderGPT - This addon allows you to use Blender with natural language commands using OpenAI's GPT-3.5/GPT-4 | Andrew Ng doesn't like it and I agree
- perplexity.ai | Question -> answer chatbot
- Today we're releasing GPT4All, an assistant-style chatbot distilled from 430k GPT-3.5-Turbo outputs that you can run on your laptop.
- OpenAI Cookbook | Examples of how to use the openAI APIs
- GenAI: generative AI tooling for IPython | Use openAI api directly in notebooks...
Podcasts
- #367 – Sam Altman: OpenAI CEO on GPT-4, ChatGPT, and the Future of AI (Lex Fridman Podcast) | Pretty solid
- pytest tips and tricks for better testing (Talk Python To Me) | Useful tips
- Episode 83: Influencers (Hotel Bar Sessions) | Fun episode, but influencers are maybe a too easy target
- Episode 88: ChatGPT (Hotel Bar Sessions) | Broke my rule not to listen to non-expert people talking about a topic I know something about -> instant regret.
- ChatGPT, GPT4 hype, and Building LLM-native products — with Logan Kilpatrick of OpenAI (Latent Space Podcast)
- Chelsea Finn joins Host Pieter Abbeel to discuss distribution shift, meta-learning, editing LLMs, single-life RL, and what can AI not (yet) do today? (The Robot Brains)
- Alexandr Wang joins Host Pieter Abbeel to discuss data, labeling, foundation models, LLMs, truthfulness, reinforcement learning for human feedback (RLHF), and AI for defense (The Robot Brains) | Nice episode slug btw 😂
- How Django Really Works - Chaim Kirby (Django Chat) | Good! How do we spread the word about Django?
- Prompt Engineering and AI Constitutions with Stephen Wolfram | Nice one