I don't understand how data centres in space make any sense at all given the heat constraints. Has there been some plausible breakthrough in heat dissipation technology recently?
Radiative cooling isn't as big of an impediment as people think. It's part of the reason why a desert that's 120F+ during the day gets so cold so fast unless there's cloud cover to radiate back.[1] IIUC correctly, the theoretical maximum is over 400 watts per square meter, and existing tech can achieve 80-90+% of that. It's an engineering challenge, but no more than creating reusable rockets. The real challenge is building a reliable convective system from the heat source to the radiating mechanism(s), and just the general problems of building a compact package for delivery that then unpacks and orients itself once in orbit. No new science is required, and you can pencil things out today the same way Musk did when penciling out the economics of building a reusable rocket.
Another way to think about it is whether orbiting data centers can be solar powered. Few people seem to think they can't. The amount of area you need to dissipate the incoming solar energy is approximately similar to what you need to capture it, so if you can do the later you can do the former.
Is it actually financially viable at all? And if so what fraction of data center build-out is sustainable at the increased marginal costs? I dunno. I agree there's probably some irrational exuberance involved in analyses. But don't underestimate the time value of money, marginal cost structures, and related financing and RoI dynamics.
[1] Convection effects are also involved, but that cuts both ways as the atmosphere is also radiating back at you, which is much less of a problem in space. A significant fraction of the net cooling is pure radiative.
No, but every day, armchair business analysts and people high on AI optimism expand the infinite frontier of blind faith and unfounded optimism in the future value of LLM infrastructure.
I wager regulatory capture lies this way. Look for this becoming an in-hardware feature, with big fat patents on it, and then a push to demand all inference providers use "TPM chipped" hardware.
Yeah as a non-American I can understand why it might be that way, but I think you should understand that globally this is not normal and is indicative of deep problems
If you're taking cameras in high schools because of school shootings as your baseline of normality, you probably need to calibrate a little differently
You sure the answer should have been 1? As a human I'd say I don't have enough information to answer this confidently, but "argument effectively" strongly suggests soft skills to me
3/10 too, and I've actually tried to discern the answer. My working theory was to pick texts which jumped between more to less frequent words (subjectively of course), but I was very wrong, basically I couldn't spot a watermarked trait at all.
It's worth noting that the bun rewrite of rust is still not actually released to users, and has run to a total cost north of a million dollars now (factoring in tokens and CI costs)
I can think of 4-5 other situations from around that era (~2012) where people were at least charged and needed a lot of help to navigate the legal proceedings to avoid jail time.
In 2010 it was more than risky on paper.
2017-2018 is well into the established era and probably even the golden age of bug bounties when a lot of corporate and judicial thinking re: white hat cybersecurity had been shifted.
reply