News for hackers. Stuff that matters.

A comment on Chatbot Software Begins to Face Fundamental Limitations. Parent comment

Re:Chatbot Software Begins to Face Fundamental Limitations by drakenot on Sunday February 02, 2025 @01:41AM (#42906565)
With DeepSeek-R1-Zero, their usage of RL didn't have reward functions really that indicated progress towards the goal afaik.

It was "correct structure, wrong answer", "correct answer", "wrong answer". This was for Math & Coding, where they could verify answers deterministically.

[ Parent | on HN ]
Re:Chatbot Software Begins to Face Fundamental Limitations by mountainriver on Sunday February 02, 2025 @07:11PM (#42913276)
It is a reward function it’s just a deterministic one. Reward models are often hacked preventing real reasoning from being discovered
[ Parent | on HN ]