Raw LLM Responses

No offense but you need to take a step back here. I use AI and chat models fine-tuned to dilute NL input into keywords. These keywords can then be used in SQL queries to retrieve set data from a data set. This takes advantage of the definitively effective NLP tech without worrying about the AI making a spelling error. You are asking a chat bot what it thinks of Star Wars. There is no way that is productive and you seem to understand that pretty well. So what’s your point? No one who is taking this stuff seriously (and believe me, I mean other, smarter people - I’m a graphic designer with a hobby) is in any way concerned about what this robot can do to act human, or what it ‘knows’, or what it can ‘reveal’. It’s a curator of the written works of humanity which is incredibly useful. Asking it opinion based questions and analyzing the results doesn’t prove anything, and it’s what even you’re insisting on doing here. When you operate it that way, sure, mimicry is a fair descriptor, but that’s really only if you abuse it. Additionally, all the people who are trying to ‘listen’ to the robot are being incredibly helpful by genuinely interacting with the model in a human-like way. The more people do that, the more data OpenAI has to work with.