Matt

[Matt @matteyas]

@whysean tl;dr: there are many intricate ideas with decades of work (the blog describes one). the blog argument is to scoff at them, then simply state (after spending some time to establish authority) that the models are already aligned and all misalignment is intentionally created. :)

[Matt @matteyas]

@QiaochuYuan perhaps maths have gone too far into purity (for public funding), who knows.

i wouldn't care too much about someone going to the other extreme of "results only." there are good reasons to do things for their own sake, even for the result-driven.

[Matt @matteyas]

@LensScientific molecular manufacturing aka programmable reality

[Matt @matteyas]

@allTheYud Can't tell with any certainty that my LLM did better, but these terms, after some light reading, do at least appear relevant from my underinformed perspective:
Fixity of the independent (direct "upgrade" to fixity of the past)
Backtracking counterfactuals

[Matt @matteyas]

@lumpenspace i love experiencing denial of experiencing

[Matt @matteyas]

@XorDev and to those who insist, at least cram some more data in there with https://t.co/ZLWUriIIAv

[Matt @matteyas]

@skdh hmm, i'm not involved in the field so i'll trust your judgment on that! :) cheers for talking about the experiment, it is an interesting one.

[Matt @matteyas]

@MathMatize *math spittle flying vigorously* that's just a model!

7 ≝ s(s(s(s(s(s(s(0)))))))

[Matt @matteyas]

@suchnerve yet another explanation:
consider S = {0.9, 0.99, 0.999, …}

taking larger members s from S, we can get as close to 1 as we like without actually equaling 1.

0.999… is bigger than any s, so it's closer than "as close as possible without equaling 1."

i.e. it actually equals 1.

[Matt @matteyas]

@Rahul151304 @grok @bungire_obang @voided @AMAZlNGNATURE because it came from space. it has travelled the universe. museums value them. researchers value them. collectors value them. dreamers value them.

regarding other stuff than iron, there are only trace amounts of anything rare. there's value in the potential of finding something.