Emerging reasoning with reinforcement learning | Hacker News
news.ycombinator.com
Emerging reasoning with reinforcement learning | Hacker News
This report is long but very good.
“With R1, DeepSeek essentially cracked one of the holy grails of AI: getting models to reason step-by-step without relying on massive supervised datasets. Their DeepSeek-R1-Zero experiment showed something remarkable: using pure reinforcement learning with carefully crafted reward fun... See more
Chamath Palihapitiyax.com
DeepSeek-R1-Zero (no supervised fine-tuning) showing human-like reasoning skills in natural language just by virtue of reinforcement learning (RL). https://t.co/P0wCC3wVtN
This "Aha moment" in the DeepSeek-R1 paper is huge:
Pure reinforcement learning (RL) enables an LLM to automatically learn to think and reflect.
This challenges the prior belief that replicating OpenAI's o1 reasoning models requires extensive CoT data. It turns out you just need to give it t... See more