Conversation

Initial results with process supervision — training an AI by rewarding it for coming up with a correct thought process, not just outputting the right answer. Great results in mathematical reasoning:
Quote Tweet
We trained an AI using process supervision — rewarding the thought process rather than the outcome — to achieve new state-of-art in mathematical reasoning. Encouraging sign for alignment of advanced AIs: …openai.com/research/impro
That's fantastic! Sounds like your AI is not just a one trick pony, but also knows how to properly get there. Keep up the process of process supervision!
1
1
Show replies
This is like the old saying about the most important thing in school, "learning how to think."
4
ChatGPT LLM is super weak in math . Today I tried to solve a math problem from 4th grade math book . ChatGPT answered wrong answer . I checked my answer with published answer as well with perplexity AI,google as well 4th grade teacher to be sure . Everybody confirmed that Open AI… Show more
1
Impressive progress with process supervision AI in mathematical reasoning! This approach of rewarding thought process instead of just output will lead to more versatile and creative AI in the future. Looking forward to seeing further developments! #AI #processsupervision
1