

That’s a lot of misplaced trust you’re putting into graduate school admins.


That’s a lot of misplaced trust you’re putting into graduate school admins.


That’s probably exactly what they did, deployed a sbc + some off the shelf model and increased the price 300x by making a black plastic shell.


Well you see they both end in -berg and fatberg is a bad thing


I don’t remember if supernote is OSS, but I’ve got the A6, and I think they got an A5 and A4 sized one. Pricey though.


I got 16 gb on my W11 laptop, and I’m regularly bouncing off the 13 gb point where swap/ compression/ whatever kicks in and shit gets slow, doing relatively benign shit like browsing the web, YouTube and Gmail open, and vscode running working on some python shit. On 8 gb I’d probably have to use my phone for everything except vscode to get it to fit. It’s not the lightest possible workload, but it shouldn’t be slamming swap so regularly.


The way eink display prices scale is insane. You can get a 13 inch one for like 200, but any bigger than that it skyrockets. Must be that they still struggle to get good display yields across that large of an area, defect rates or something.


If your argument was one of cost, you should have said so from the start! Economically, it might or might not make sense. I can’t pretend to know the economics of running a space based datacenter, I’ve never run a ground based datacenter.
But you have been arguing about power and electricity and heat and how proud you are to have 200a service at your house (congrats on owning it, btw, tough nowadays) but those aren’t the dealbreakers. If the AI bros want to lose billions putting the datacenters in space, I don’t have a huge problem with that. Better that than diddling kids and destroying society, which is what they seem to be spending their money on now.


You don’t use 48 kW you have 48kW capacity, that’d be 33 (1500W) electric space heaters running nonstop 24/7. I have electric heat, electric oven/range, and an electric car and I averaged 3 kW across the last week. (406 kWh between the 26th and 1st)
A comparison that is reasonable is an h100 rack cluster like this which uses about 60 kW per rack. For input power, the newer iROSA solar panels generate about 20 kW at a size of 20ft x 60ft each. Throw in 4 of those radiators, and you have something that is feasible to throw into space. Again, I can’t judge the economics of launching and running a space based datacenter business, but you could absolutely launch and operate a space rack with current tech.


I could have seen the opposite, too, where xAI pays spaceX gobs of money for tons of datacenter launches in advance, bubble pops, xAI goes under and spaceX still has the money. But ya I think this is “the govt won’t let spaceX fail, pile all the riskiest shit into that”


That was my stumbling block, too. Don’t think of it as taking a datacenter and putting it into space whole, think of it as taking 5 or 10 racks and putting that into space, and repeating till you have as much compute as a datacenter. So it’s basically the size of a schoolbus (same size as hubble telescope) and it has solar panels+ heat rejection like those of the ISS, and then bolt a starlink on the end, and you can put as many of those in orbit as you need.
Each part of the hardware is doable(ish), and if the nerds who actually run datacenters say the terrestrial energy/cooling cost numbers vs launch cost numbers make sense, I’m inclined to believe them even if I don’t get to see that math specifically. But right now it’s just AI bros saying the costs make sense, and I don’t as much believe them.


I used to think space data centers was a scam, but I learned how much power existing satellites already use (and thus must be able to radiate into space to keep cold), and just looking at the ISS, each radiator (and it has several) can reject 14 kW into space, so if the ISS has can safely generate 14 kW of electricity and reject all the waste heat, then the major concern for me is addressed. Space datacenters are the first step to industry, in space, which is an necessary step for a lot of future stuff.
All the above is beyond the point though, he’s playing shell games to tie his most valuable and critical company, SpaceX, to the trashheap of AI bullshit so that the government will bail him out when it crashes.


It’ll be more like ordering a Chinese “first person RC car” then grabbing a conversion kit. this company newbeedrone currently sells a drone, and a kit to convert it to a car

But instead they will reverse it. You get a cart, and they sell you a $20 drone conversion kit personal fan add on


We’ve tried nothing and we’re all our of ideas!
Gtfo, if you want to eliminate crunch, just don’t have them work more than 40 hours. It’s easy accept that youre gonna miss some deadlines, cause you fucked up the planning. Promise your workers triple pay during OT and suddenly the middle managers ain’t as motivated to try to force unofficial OT


Wait until he learns that physics is just theoretical chemistry
Definitely. It might require significantly more input to gain the same certainty, but all it’s doing is reweighting the possible next tokens before choosing, and code output is still just token output. The rigid syntaxes probably means that the next token probabilities are much more sharply divided (maybe a random sentence the top 1 choice is just 40%, top 3 are 80%, but for a line of code, the top 1 choice might be 90% probability and top 3 hit 99%)