My proudest is definitely David Holz. Self-funded a company with over 500M ARR. With good vibe on both art and research, insightful on market and product, in-person host weekly meetings with users, real legend.

My mental / physical health stack - 7hrs sleep - Meditation - Long walks with friends - High flow shower head - Daily Bryan Johnson vitamins - Daily Bryan Johnson longevity mix - 2x weekly sauna + hot tub - 5x week Tonal Fitness workout - Pre-portioned healthy meal prep

I find this new paper aligns with my idea and thoughts in my previous post, it's better reasoning models <-> better reward models. It's even self-evolved from a rule-based environment, insightful. arxiv.org/pdf/2505.14674

interesting reasoning model seems better at more than expected

It's interesting to see every RL paper would write down GRPO formula even they just apply it to certain domains or show some empirical results. Similar vibe to the trend that when 2020 people are applying transformers everywhere.

I’ve worked on evals github.com/EvolvingLMMs-L…, and once considered turning it into a business. But here’s my honest take on the future of evaluation: Evaluation should be non-profit and public. People naturally trust evaluations from non-profit organizations more than from

Elegance of co-design!

I’ve been using Claude Code by Anthropic and it’s been a massive productivity boost — I now complete tasks 12x faster. For example, integrating a new eval task into LMMs-Eval used to take me 2–3 hours (even with Cursor), but now I can do it in just 15 minutes. When I added

Anthropic's CEO claims AI hallucinates less than humans. Bold statement. So I decided to test it by feeding the same FAKE theories to ChatGPT, Claude, and Gemini to see which one calls me out first. The results shocked me 🧵

Does this mean self-consistency is enough for model to develop to better state?