@aven mixed but I got some interesting results. It's definitely not reasoning like a person, just kind of saying what it thinks you expect. it's more for breaking down what the AI is really doing than exploring racism or its motivations/causes. I don't think you could use it to make a real-world point about it.
@a7@Moon that's an interesting insight. I think there are definitely people who are genuinely racist as a matter of principle and genuine belief, but a lot of them are social chameleons that just "when in Rome" with agreeableness, and will be racist or not-racist depending on what is socially acceptable.
It would make sense if agreeableness and outgroup-hate were correlated. Like they don't actually have an opinion, but just want to get along and be popular.
Also, maybe more to your point, there are covert racists who just don't want to get into it if there's a high social cost to speaking their true opinion. I wonder whether that can be simulated with AI.
@Moon@shitposter.club@aven@shitposter.club this is how people act irl, really racist people irl will feel you out to see if you are down with hating whatever group they hate and then if you don't signal you agree with it they will fold and start faking opinions that please you.
@Moon@lain@aven The discount way is trying to fit the book into the context (maybe using summarization or only opening a window into the relevant info with a vector database). The proper way is fine-tuning/LORAs/QLORAs.