A new attack has been developed that can extract hidden architectural details from GPT-4, PaLM, and other models using just API access and $20.

1m read time From notes.aimodels.fyi
Post cover image

Sort: