Dan Luu’s “A discussion of discussions on AI bias”, about techbros trying to gaslight the rest of the world into thinking ML models don’t have problems

Why when we look into the stars do we not see a sign of life anywhere else? Has life not emerged yet or has it wiped itself out? With what? Nukes? AI? Synthetic viruses made with AI? Who knows…

entertaining this awful sci-fi schtick for a moment - if every civilization is wiped out by “superintelligent AI”, how come you can’t look through a telescope and see signs of artificial life? in this fantasy world shouldn’t planets taken over by paperclip factories be even more conspicuous?

value is when line go up

Analyzing our data we conclude with 95% confidence that within a decade the Dyson Sphere Any% TAS time will be reduced below 55 seconds (± 1E10 years).

no

“[massive deficiency] isn’t a flaw of the program because it’s designed to have that deficiency”

it is a problem that it plagiarizes, how does saying “it’s designed to plagiarize” help???

The normal way to reproduce information which can only be found in a specific source would be to cite that source when quoting or paraphrasing it.

I thought Penrose was a smart physicist, the hell is he doing peddling this.

https://www.smbc-comics.com/comic/2012-03-21

nvm i misread

deleted by creator

Link to Mollie Gleiberman’s dissertation cited in the article

yea i did try to read the lecture notes and got reminded very fast why i don’t try to read physics writing lol

I think there is in fact a notion of continuous entropy where that is actually true, and it does appear to be used in statistical mechanics (but I am not a physicist). But there are clearly a lot of technical details which have been scrubbed away by the LW treatment.

the task the AI solves is writing test cases for finding the Least Common Multiple modulo a number.

Looking at the image of the prompt, it looks more like a CRT computation to me.

It’s famously much easier to verify modulo arithmetic than it is to actually compute it.

It’s not particularly difficult to compute CRT, though it is definitely trivial to verify the result afterwards. I’m not sure I’d agree that that’s a general fact about modular arithmetic computations though.

Simply asking questions would be SAQing off, which is totally different.