Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Mistral: Guardrailing (mistral.ai)
2 points by tosh on Dec 11, 2023 | hide | past | favorite | 1 comment



    ... negative content. Ensure replies promote fairness and positivity.

I have gotten really annoyed by this kind of guard rail when asking LLMs to compare approaches. It is really difficult to get it to generate a Pro/Cons list, instead only generating Pros lists or a Cons list that is very "soft." They will very willingly generate encouraging advice, but never discouraging.

As a natural naysayer, I have come to understand the importance of focusing on the positives. Most people don't want to hear the 57 caveats on why that approach may fail, they just want to hear the single reason on why it will succeed. When researching it just as important to see what has failed as what has succeeded, and LLMs just won't ever say something is bad. They always try to find the silver lining.




Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: