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

Could this be improved if the evaluation was done by an independent sub-agent?


Is it running out of space in its context window?


My rational is that perhaps it's being biased towards continuing doing what it's doing, or biased towards telling that it has done a good job and not being self-critical.




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

Search: