Pedro Domingos argues RL solves nothing that better generalization can't
pmddomingos · x · 2026-10-23
AI researcher Pedro Domingos posts a contentious take: reinforcement learning doesn't solve any problem that can't be solved by better generalization. The claim implicitly challenges the current narrative that RL training is the key driver of frontier model capabilities, framing RL's gains as fundamentally borrowed from generalization.
Related event: Pedro Domingos claims reinforcement learning is basically useless(3 posts)→
More from AGI Musings
- Bay Area tech layoffs hit nearly 14,500 in 2026, already exceeding all of 2025 — rohanpaul_ai · 2026-10-24
- Both 'rogue AI' fear and AI welfare narratives dodge corporate responsibility, argues dev — AlexTensor · 2026-10-24
- AI financial shenanigans mount as AI-generated content takes over most of the internet — michalmalewicz · 2026-10-24
- Tim O'Reilly gives Claude a byline: interview with Mozilla CTO on open source AI — latticecut · 2026-10-24
- Cloning Your Boss: Distilling a Year of Slack Into an AI Persona File — every · 2026-10-24
- genmon's singularity joke: step out for a week, come back to 1000-IQ posthuman friends — genmon · 2026-10-24