LyingGPT

I bet that "bias" on the medicare question came from the previous responses.

Every time you hit the submit button to an LLM, its like it wakes up from cryosleep with no conscious memory but an unconscious understanding of stuff, then its immediately handed a script... and the whole script is the conversation. That is its entire conscious memory.

That previous conversation gives it its personality, and I suspect it would answer the medicare question differently with a blank slate.