Using an LLM for finance is insane, not only for the usual reasons AI is disliked here on Lemmy, but because the proper way to do this (putting aside whether you should) would be to tune a machine learning model, probably a supervised regression model, to do these predictions.

Every single fucking piece of AI intellectualism seems to be :

We took everything that people have made and turned it into a statistical model. We are outraged that it's accuracy is similar to an average person, sometimes better and sometimes worse based on the weighting of the data we fed it.

Surprised pikachu face.

It's trying to find the most statistically likely response to your inputs. if it thinks the situation will statistically lead to whatever response/behaviour, it will do it. (EDIT: if the majority of the text it was fed on gambling has the person going for "one more bet, if I bet my wedding ring I can win it back" that's what the response is going to be exactly that)

"Why does 'pretty woman' come up with only white women?" because the data set human beings created or fed into it had a disproportionate amount of images tagged of pretty white women and less pictures tagged of pretty black women. It's the fucking data set, it's a statistical , non deterministic, heavily biased machine. Of course it's giving out biased BS.

EDIT : To be clear, if they fed the machine more pictures of pretty black women and less pictures of pretty white women, it would be biased towards generating pretty black women if you prompted it to show you a picture of a pretty woman. I am not racist, I am just observing racist things happening and pointing them out.

Only if you're doing it right.