Dwarkesh Patel
4 mentions across all digests
Dwarkesh Patel is a podcaster and AI analyst who has interviewed Anthropic CEO Dario Amodei and Microsoft CEO Satya Nadella, and published analysis arguing RL is more information-inefficient than commonly understood due to sparse reward signals.
What I've been thinking about this weekend - More open questions, intelligence vs power, the problem of verification in science, the parallel discovery of Darwinism
Five hyperscalers control 70%+ of global AI compute with most concentrated at OpenAI, Anthropic, and Google DeepMind — raising questions about whether this concentration prioritizes frontier research over empowering ordinary users and risks pricing the masses out of AI entirely.
Blog prize for big questions about AI
Dwarkesh Patel's $20k blog prize on AI research fundamentals—RL scaling, training dynamics, compute, US-China competition, governance—is a talent recruitment play to hire a research collaborator by May 10.
Thoughts on AI progress (Dec 2025)
RL is even more information inefficient than you thought