@ThunderousProphet Do we humans do any better?

I'm comparing this to "train of thought" - like how I am often going down a route of discovery. And it is only further "down the line" where I realize if it is a dead end or not.

Isn't that pretty much exactly how AI, as in predicting the next word, is sort of working?

We just dummify it because the AI is exposing its "train of thought" to us as it goes.

That is fixable!

@tsvenson @ThunderousProphet I'm still concerned that we're censoring the outputs when they're scary to us rather than dealing with root causes.

If the prompt would lead to sex, drugs or violence, ChatGPT generally refuses to answer. Then the hackers get in there and with a little clever rhetoric expose that it's perfectly capable of concocting plans to commit a crime, take over the world or build a bomb, we just handicap it from doing so openly.