Thoughts — Jason Wei
Levels and expectations Bet on AI research experiments History of Flan-2 When I don’t sleep enough Thinking about history makes me appreciate AI Advice from Bryan Johnson Sora is like GPT-2 for video generation A typical day at OpenAI Yolo runs Uniform information density for CoT Inertia bias in AI research Compute-bound, not headcount-bound Magic of language models Why you should write tests Co-founders who still write code 2023 Hyung Won Read informal write-ups Relationship board of directors Reinventing myself Good prompting techniques 10k citations Manually inspect data Language model evals Amusing nuggets from being an AI resident When to use task-specific models Benefits of pair programming Many great managers do IC work Why I’m 100% transparent with my manager My girlfriend is a reward model Better citation metrics than h-index My strengths are communication and prioritization Emergence (dunk on Yann LeCun) UX for researchers My refusal The evolution of prompt engineering Prompt
2026 A cognitive core is not enough 2025 AI will not takeoff fast Work in the office AI should create unicorns The asymmetry of AI research Fact-checking good AI researchers The description-execution gap RL environment specs The inverse 80-20 rule Method-driven vs problem-driven research AI research is a max-performance domain ATLA is the ultimate benchmark Measurement is all you need AlphaEvolve is thought-provoking Binary-choice questions for AI research taste The craziest chain-of-thought The best hard-to-solve easy-to-verify benchmark Flavors of AI for scientific…
saved by
related reading
- LessWronglesswrong.com
- My picture of the present in AI — LessWronglesswrong.com
- Tips for Empirical Alignment Research — AI Alignment Forumalignmentforum.org
- The Universe from an Intentional Stancecasparoesterheld.com
- blog/2024.09.impact.md at main · okhat/blog · GitHubgithub.com
- As Rocks May Think | Eric Jangevjang.com
- Harvard CS197: AI Research Experiencescs197.seas.harvard.edu
- Sara Fish • LLM Evals Research Advicesara-fish.github.io
- Sara Fish • LLM Evals Research Advicesarafish.com
- Ai2: Truly open breakthrough AIallenai.org
- Zen and the Art of Machine Learning Researchblog.jxmo.io
- LLMs Go To Confession, Automated Scientific Research, What Copilot Users Want, and more...deeplearning.ai