Lessons from Isaac Asimov on Taming AI

You've really got to laugh/cry at someone who looks at the utterly milquetoast responses given by ChatGPT and everyone including investors collectively losing their shit and says ... "how can I go from owning small% of this company on the hype train to 100% of a company on this hype train".

"Wont somebody think of the children, its not safe" is the new "Creative differences" as the excuse for breaking up the band and going solo.

It's not a question of being edgy[0], it's that we don't have the faintest idea what it even means for an AI to be aligned with our values.

Closest we have now is following the intuitive ideas that generally work for most humans — giving positive/negative feedback for particularly good/bad answers, and asking nicely in plain language — but last I checked, the people who have made it their mission to seriously consider what it means from first principles consider this to be so unsophisticated they call it "pre-paradigmatic"[1].

[0] unless you're worried about minors getting access and learning what the swear words mean, which is a thing

[1] which was a bit over a month ago, so practically lifetime in AI: https://www.planned-obsolescence.org/disagreement-in-alignme...

That is with "guardrails".