I've noticed the ChatGPT is much better at gathering "deep information" in the m... | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

TheCaptain4815 on Nov 30, 2022 | parent | context | favorite | on: ChatGPT: Optimizing language models for dialogue

I've noticed the ChatGPT is much better at gathering "deep information" in the model compared to 003-instruct. While this could be solved with better prompt engineering on the 003 model, giving the same prompt to ChatGPT has given me much better results.

make3 on Nov 30, 2022 [–]

apparently (according to the blog post) that's a result of the RL human preference fine-tuning - the human rankers preferred longer more in-depth answers

TheCaptain4815 on Nov 30, 2022 | [–]

Wasn't the instruct model created using that same strategy?

Consider applying for YC's Winter 2027 batch! Applications are open till November 2.
Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact