ChatGPT jumped into analyzing my personality based on the system prompt, and when challenged about the behavior it should have access to being only the one line question, it prevaricated.
Claude looked for the directory of transcripts, found it empty, told me it couldn’t answer the prompt given current information, and offered a few ways forward if I wanted to get an answer to the prompt.
https://web.archive.org/web/20260213222727/https://old.reddi...
I thought this was an interesting concept. It feels a little dystopian in a corporate context but much less so in an early stage startup.