@haise yes. I tested now my local AI with feeding all kind of natural science textbooks and peer reviewed papers. It seems really working great. Now I try soft science like psychology, sociology etc. where you can't know if they deliver hard facts or just some ideology. Soft science will be more difficult to get results you can count on. But it will be interesting to see what AI will make out of it. AI doesn't really know what it is about. Our AI is just a language model to simulate knowledge as if it is conscious of. But the AI is not really aware of the context it is generating. It could be more interesting when we get AGI that is capable even to make own analysis because if can also do mathematics by itself and use different models to adapt it for various fields. But we are still far from AGI. Perhaps we get AGI in 2 years.

@haise what's also important is to have many sources. When you get answers just from one source then you have to look carefully. But if there are 3 sources as textbooks you can look up then it seems to be good answer. When I try the test with soft science I try to get many textbooks as well and see then how this works.