New project to compile a Q&A dataset of ChatGPT errors crowd-sourced from social media. Don't know of any dataset assembled adversarially like this against an LLM — feels valuable for future model comparisons:

→ View original post on X — @goodside