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

Over the long term LLMs are going to become very interesting snapshots of history. Imagine prompting an LLM from 2025 in 2125.


I would probably prefer wikipedia snapshots (including debate) as a future historian.


The more options you have, the better IMO.


You're right: I wish OpenAI could find a way to "donate" GPT-2 or GPT-3 to the CHM, or some open archive.

I feel like that generation of models was around the point where we were getting pleasantly surprised by the behaviors of models. (I think people were having fun translating things into sonnets back then?)


Maybe in the sense that a CueCat is interesting to us today.




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

Search: