@NVIDIA AI Scientist. @Stanford PhD. Building embodied general intelligence. Sharing hot ideas & deep dives🧵! NeurIPS Best Paper: MineDojo. Ex-Google, OpenAI
Featured content
See AllRecent recommendations
See AllRecent posts
See AllThis is by far my favorite post on the AI executive order. Andrew says it best: The right place to regulate AI is at the APPLICATION layer. Requiring AI apps like healthcare, self-driving, etc. to meet stringent requirements, even pass audits, can ensure safety. But adding burdens to foundation model development unnecessarily slows down AI’s progress.
Great minds think alike! Awesome work on your end too!
Everyone should read the celebrated mathematician Terence Tao's blog on LLM. He predicts that AI will be a trustworthy co-author in mathematical research by 2026, when combined with search and symbolic math tools. I believe math will be the first scientific discipline to see major breakthroughs enabled by AI, because math: â–¸ can be expressed conveniently as a coding problem. Strings are naturally first-class citizens. â–¸ can be rigorously verified by theorem provers like Lean, rather than relying on empirical results. â–¸ does not require physical experiments like biology & medicine. Robotics isn't ready yet. We are already seeing big progress: â–¸ LeanDojo from my colleagues @NVIDIAAI & @Caltech is among the first steps towards this grand challenge. â–¸ Last year, OpenAI used Lean to solve some math olympiad problems: â–¸ ChemCrow is another example, but for chemistry. It integrates GPT-4 with professional tools like molecular synthesis planner and reaction prediction: â–¸Terrance Tao's blog:
A well-written thread by Shawn on the trend of GPT becoming a proficient tool *maker*, in addition to the usual tool user mode. We used to design a shell layer on top of LLMs, but now LLMs can act as a shell "controller" on top of codebases they implement on the fly. Nice…