Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Reflection Llama-3.1 70B: Testing and Summary of What We Know (datacamp.com)
2 points by alexolteanu on Sept 9, 2024 | hide | past | favorite | 1 comment


> The model incorrectly concluded that 9.11 is greater than 9.9. However, the creator suggested adding “Think carefully” to improve the model’s accuracy.

Sorry, but if you ask a binary question and model responses incorrectly, model failed. If after receiving incorrect answer to a binary question authors suggest to add "think carefully" or "please change the answer to less intuitive" or other switch they baked into model, do not fall for their persuasion, as this is a direct deception and fraud.




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

Search: