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

I've had better results separating these concerns rather than trying to stuff it all into one prompt. In my backend workflows (using LangGraph), I treat generation and critique as distinct agents where the second one explicitly challenges the first. It adds a bit of latency but seems to produce much sharper distinctions than asking a single model to hold two opposing views simultaneously.




Ah interesting. i like actor-critic models! do you use it just for coding or non-technical chats too?



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

Search: