niplav @niplav@schelling.pt

@BasementRat@toot.site
glad I picked easier goals

@genmaicha which option do I pick if I'm not a cis woman

@TetraspaceGrouping I mean *what were they thinking*. Did they not know what was coming for them. Did they really expect EY to talk about ChatGPT or what.

Why did EY go on that one. The whole NFT framing feels kind of scammy, and e.g. the 80k hours podcast would've hosted him I'm sure.

So many questions.

@alexthecamel not surprised given the context

@WomanCorn awwww cute

Ah why do I even honest-reply to people who are just interested in performative hate

@sundogplanets

Ah to be clear on the policy level I absolutely think that we should Harberger-tax the orbits around the earth and Pigou-tax releasing material into the atmosphere by deorbiting satellites.

@sundogplanets Ah, I see. Is aluminum more harmful to the atmosphere or people than silicates?

Skimming Wikipedia doesn't give me anything that sounds dangerous (some slight environment effects such as inhibiting gill-breathing in acidic water and plant-growth on acidic soil).

I wonder how much aluminum ground-based entities emit into the atmosphere.

Nobody's gonna do it if it doesn't result in money, status or sex.

(Lots of things don't get done even if they result in sex! Surprising)

@sundogplanets Unless their material is radically different & more harmful maybe?

@sundogplanets
Hm.The wikipedia page on meteoroids states that "An estimated 25 million meteoroids, micrometeoroids and other space debris enter Earth's atmosphere each day, which results in an estimated 15,000 tonnes of that material entering the atmosphere each year."

If I'm not mistaken, this would make another 29*365=10585 tons of matter entering the atmosphere per year. Perhaps this has an effect, but it's certainly not completely new.

q/acc

qualia accelerationism

more towards zero than one (so <<5%). Obviously the value-laden part wouldn't be solved.

I think the philosophy/math/cs system would be just as capable at capabilities work as at alignment work.

But I now remember an old idea of making STEMGPT, trained (in the weak case) only on STEM textbooks, arXiv, (in the strong case only on) hadron collider data, protein structures, meteorological and geological data &c. Hard to have info about humans leak over though.

How much of strawberry alignment is value-laden? 5%? 95%? probably further along some logarithmic scale, if I had to bet

The, ah, fifth thing I disagree with Eliezer about.

Even with ML systems!

I agree that probably with most architectures, if you train them a lot to be capable alignment theorists, they have inner optimizers that are capable consequentialists, but the alignment-theorist-phase might be quite long (I could_{10%} see it going over 100x human ability).

If we had those widely distributed, people would likely use them for capabilities and just widen the gap (e.g. OpenAI who talk about this as a strategy are not to be trusted with that strategy, since I don't see them using it solely for alignment work for half a year, and instead using it on both capabilities and alignment. But their plan is sound in that regard).

But I disagree with the view that you can't have the alignment theorist that is not also a consequentialist.