logoalt Hacker News

throwaway4aday01/22/20250 repliesview on HN

That's essentially what R1 Zero is showing:

> Notably, it is the first open research to validate that reasoning capabilities of LLMs can be incentivized purely through RL, without the need for SFT.