https://www.youtube.com/watch?v=lk3jCuITwcE
TLDR: If you give a chatbot a few recursive instructions, it will start using a large number of tokens and hallucinating. I wonder if this applies to all LLMs or if it's just the cheap scammer versions. If it works for all LLMs, it might be this generation of AI's version of "This sentence is false".
[link] [comments]