Has it *really*? From what I've seen they have most of the specific tasks to the point where they can be highly automated for normal operations, up to and including now a mechanism to select an airport and radio announce the emergency landing and land the plane autonomously.

The humans are needed for exceptional cases, and that isn't about to change even if theoretically the plane is theoretically capable of autonomous flight.

Regardless of the merit of ChatGPT. My impression is that Altman isn't a contributor to it, but gets all credit by his position above it.

The difference between Altman and someone like Milton is that Altman lucked out to have the technical team that essentially "won the lottery".

It's a fairly common pattern, someone purports to be a "thought leader" with pretty obvious goals with no inkling how to get there, and if their pet technical people pull it off, the comparatively useless guy gets all the credit and the actual people pulling it off remain comparatively anonymous.

Without lucking into others' technical success, they are more likely to be known as a high profile grifter if they are known at all.

The amount of money in play may dramatically reduce, making it less appealing to people only in it for the money.

If there's anything that can lose out to AI slop, it's commercial software slop. An ecosystem of vendors actively looking to screw over their customers with no talent hacks at super high prices, which will AI slop themselves up to get the worst of all worlds.

Few business will want to pay for "better" because the suits believe as hard in AI slop as they have believed that saying the word 'Agile' magically fixes all that ails their processes. So software development might become a respectable, but modest way to earn a living and scare away the high rollers.

The size of the ISS includes both the radiators and the solar panels, so absolutely that is the scale of an ISS unless you ignore the solar altogether, which you seemed to pivot toward "footprint doesn't matter".

Even if you quadrupled the scale to compensate for 50% loss of solar and idling datacenters if you didn't want an energy storage solution, I would suspect it would still be more cost effective in aggregate. Without making tons of space junk and every thing necessarily be single use components to burn up on re-entry.. if we are lucky.

Whatever the future of this, space is a dumb place for it. The starlink constellations are annoying enough and this would just be amazingly worse and a waste of our resources.

Also remember, this won't be just a handful of potential deorbiting disposable nightmares, it will be tens or hundreds of thousands if they actually executed the way they want.

The fact remains, something the size of the ISS would be needed to support ~100kw of space-hosted stuff. The radiators are only "tiny" by comparison to the big solar panels, but they are plenty big, and we can't ignore the solar footprint that would be needed.

We should not have tens of thousands of ISS magnitude things in LEO to feed the 'wouldn't space be neat?' hubris of a handful of self-congratulatory "thought leaders".

Though think of the opportunity for nVidia, if they get everyone sold on the 'your GPU will burn up on re-entry in 5 years', then they never have to worry about a client delaying a refresh...

The ISS power/thermal solution is good for about 100KW, not a MW.

Power: We have plenty of power now, and as was stated, the panel sizes are still huge for the power needed, even in space
Launch cost: Doesn't matter who 'owns' it, the launches are still expensive, regardless of how you move the shell game of accounting
Latency is still at a disadvantage compared to a boring old fiber, regardless of your fondness for Starlink
There is *zero* reason to be concerned about a common solution for Moon and Mars at this point. Even if there were, landing on those rocks provides a much easier heat sink than having to bring the whole sink with us.

Yes, that's the whole problem, "big enough" is just massively big, and you end up with a *whole* lot of overhead cluttering LEO and a whole lot of waste launching it...

And the *only* rationalization offered so far has been "oh, this one company that makes a part for natural gas power plants is backordered". There is zero chance that the *easiest* solution for that is tens of thousands of ISS sized datacenters.

Yes "enough radiators" means something the size of the ISS for a couple of GB300 grade servers.

The possibility isn't the thing, it's the sheer impossible scale is ridiculous compared to "just do it on the surface, where you have a nice convenient whole planet to sink the heat into".

The math is that for each couple of GPU servers (we are talking GB300 sorts of servers), you'd need something about the size of the ISS' solar and cooling systems. Contrary to his "great engineers will solve it" sentiment, the cooling side of the equation is governed by very straightforward and incontrovertible physics, best possible solar isn't *much* better than what the ISS has already.

Others in this "let's go orbital datacenters" crowd have already pegged that you'd need tens of thousands of these setups, meaning that we'd need to launch essentially tens of thousands of ISS', and probably plan to let them all burn up after 5-6 years.

This is to compete with maybe 3 to 5 traditionally terrestial datacenters, the construction of which would probably be less than launching even a handful of those setups.

Yes, *technically* possible but physics is why the economics is completely stupid. Given enough materials and devoting an ISS-scale power and cooling solution per server we can work, but it's monumentally stupid because ultimately it's a massive waste of resources for a highly compromised theoretical end game versus just plopping them down in terrestrial datacenters. The physics involved in the data transmission, the power, the cooling, the launching is all very stupid, possible, but stupid given the huge cost versus much more straightforward alternative solutions.

The 'we got lots of great engineers' says "we executives have provided the thought leadership and now the engineers better deliver on our crackpot idea or else it's all *their* fault that they couldn't deliver our brilliant idea"

Given that these folks withdrew their funding upon finding out how it was being used, it's self evident that these people at *least* did not want to be paying for machine translation. Maybe they expected just acquiring and scanning materials or human translation, but machine translation actively drove them off. We don't have to speculate on whether these folks minded their money to go towards LLM translation, they said very plainly.

In fact, "GPUs" are evidently whatever nVidia wants them to be.

nVidia has been obnoxiously pushing the narrative that they invented the very first GPU ever in 1999. Despite not being the first or even the first to use the acronym, but their marketing picked GPU to describe the Geforce 256 and obnoxiously the marketing groups selection of an acronym that Sony used earlier is the biggest thing retained in their "history" from that era.