> Maybe I'm just weird, but I love my wife and wouldn't cheat on her
I agree with you, but, brain chemistry is weird, and "wouldn't cheat on them" is easier said than done and requires a lot of discipline in the heat of the moment when the situation arises and your own hind-brain and hormones are fighting you. A blasé attitude won't prepare you foe the sucker-punch.
I've seen grown people with multi-decade relationships throw it away because they got a tiny bit of extra-marital romantic attention, and the thrill regressed them to teenagers. In the majority if cases the cheaters' amors weren't interested in a long-term relationships.
That's a valid and measured take, but most of the comments in here are not talking about the affair in such terms. They are mostly painting Jensen as some kind of innocent child who was manipulated with Jedi Mind Tricks, when there's no reason to think her approach was anything but blunt.
I don't even care that much about what Jensen did, I don't even know what his marriage is like. It wouldn't bother me if he stayed on Nvidia. But people are defending him in a way that comes off less like "we can all make a dumb mistake on the wrong day" and more like "all men want secretly to ditch their wives, and his mistake was doing it with a spy."
No. Never spent money on these sites. Public website/app. You could read any comments publicly too. Never even DMed, until ... Some things just happen, quite randomly. No more details.
I wasn't judging, I wanted to understand how it happened, and the added context is appreciated.
I don't know of that was your case, but many times, its a series of small, seemingly justifiable yet ill-advised, boundary-pushing (yet thrilling!) decisions that lead to somewhere messy.
Sure. I understand. My comment was meant to indicate that I never thought I would get where I did. Clean until you are not. Yes. Small small things becoming an obsession.
> Almost all statements Yann LeCun made back then (LLMs won't be able to do this or that) turned out to be wrong
You're being uncharitable. He would have been right if Chain of Thought hadn't been pulled out of a hat, and no one saw it coming, or how u reasonably effective it would be.
Today, no one can know if there will be a future step-change improvement to LLMs at the same level as CoT, and conversely, no one can be absolutely certein there won't be - so it shouldn't be a firable offense because neither position obvious; even though it will look like it in hindsight.
The most confusing thing to me about AI and LLM progress in the past 4 years is that I can't clearly see how all novel patterns add up, or what was the right point.
I think adding harness + MCP + chain of thought + specialized data generation resulted in the world where we have now - the one where you don't have to type code manually, but sometimes do still have to think and sometimes don't.
But LLMs without all this tooling and calls under the hood are still kind of... LLMs. If there's a space to hallucinate - they will hallucinate.
CoT wasn't pulled out of a hat. It was obvious to any researcher that it was needed. It just took some time and stronger base models and longer context to actually be implemented. People were already doing "CoT" by telling models to think step by step before answering, primitive harnesses and similar tricks.
Even a nobody like me was playing with a sort of CoT with llama finetunes before o1 came out.
Only this year I'd say we picked all the low hanging fruit that has been obvious for a while that was needed.
I'm not fully up to date on AI research anymore, however I think the obvious stuff has already been implemented, but there's still a lot of room for improvement and scaling for these current approaches.
For verifiable tasks like coding and math, you can keep going, doing more and more RL. You have the AI generate hard problems, then try to solve them and reward correct solutions. We basically have RSI for verifiable tasks, we just need to keep the wheel turning.
> They may be ugly, but the emergency prepper in me always wanted one after hearing that story
If all you're after is the V2L feature, you can get it for much cheaper: for 16% of the price of a Cybertruck, you can get 86% of the battery capacity (about 8 powerwalls) by buying a used Fisker Ocean Extreme with a 107kWh battery, vs. the cybertruck's 123kWh
> FWIW, given how many places in the U.S. are passing datacenter bans
If you have the technical chops to set up a data center in space — with no connections to the electrical grid or municipal water, you can use your technology to build your data center with the same specs right here on earth in 90% of places that you think are issuing "bans". You'll also get much better cooling for free via convection.
A remote town with a population of 43 will be happy for a data center to employ 2 locals, but data center operators aren't doing that today because they want tax breaks tiny towns can't give, infrastructure tiny towns don't have, nevermind a well-armed police force to deter would-be GPU thieves. Space is a red herring.
If all the energy for a space appliance comes from the sun, in steady state they only have to radiate away the same amount that hits them. All long-lived spacecraft manage it.
Whether they can be worth the money it cost to put them in orbit is another story.
Apples to oranges, long-lived satellites have modest power needs. Doing a back-of-the napkin math on required size of solar array and radiative elements for a single-rack's worth of energy is a sobering exercise for those who are a overly exuberant on the topic. The all-in power estimate for a single GB300 chip is 2.1kW
Well, it can’t use more than it can get from the sun that hits it. So it can’t need to radiate more than that away. So it’s radiator size needs are in proportion to its solar panel size needs, and no worse than that.
Edit: ISS is about 90KW steady state, and the collector to radiator area ratio is about 5:1. Dunno about the mass ratio.
My point is not that space data centers are viable or sensible. Only that the need for cooling is limited. If a terrestrial datacenter was powered only by its rooftop solar panels, cooling it would be trivial. Space is harder, but there just isn’t that much heat to dump.
A simple rock in orbit radiates exactly as much heat as hits it from the sun, on average.
> Well, it can’t use more than it can get from the sun that hits it. So it can’t need to radiate more than that away
You are correct that the first law of thermodynamics dictates the energy, but anyone who has parked a car in the sun will tell you, the temperature where the absorption/dissipation equilibrium is achieved is often incompatible with the survival of the cargo.
edit to your edit:
> Edit: ISS is about 90KW steady state, and the collector to radiator area ratio is about 5:1. Dunno about the mass ratio.
Cool; that's about half the wattage of one modern full-height Nvidia AI rack (172kW) - you need two ISSes to service just 1 rack; extrapolate that to a data center's worth of racks and see if the math is mathing. The ISS is about the size of a(n American) football field.
It's this huge misconception that space is super cold. A better mind model is that space doesn't have much of a temperature at all just cause there's nothing there. Vacuum doesn't have a temperature.
You won't immediately freeze if you enter space without a suit. You'll die almost immediately, but it will be because your blood will begin to boil because of the lack of pressure kinda like an extreme form of divers sickness - it's just not the cold that'll get you first.
FYI; that's not how it works, you and the rock you mentioned in an earlier comment cannot have the same temperature of -18C. This is due to differences in material, and geometry to a lesser extent.
Even if you were rock-shaped, the rates of energy absorption and radiation is what determines your remains' temperature, this ratio differs for different materials, and thus difference fiilnal temperatures.
No, not just a bit. It varies by hundreds of degrees depending on the materials α/ε ratio.
Your mental model is off: materials that good absorbers but poor emitters will get very hot, and the flip-side stay cool, and are used to make the cooling elements in the ISS . The temperature of ball of black nickel at 1 AU will settle at balmy 239C - not -18C. Polished Aluminum -> roughly 145C.
There is the additional problem that the GPUs are not in the panels, so to radiate the heat you must first transport it from the GPUs, then distribute it uniformly over the entire surface of the radiators, otherwise some parts will overheat.
Without gravity, it is likely that this can be done only by pumping continuously a lot of liquid, which will increase the power consumption and it will reduce reliability.
Solar panels that have heat radiators on their backside will have a higher temperature than simple solar panels, where half of the absorbed solar energy is converted into electrical energy and taken away. The efficiency of solar panels decreases with temperature, so hotter solar panels will have to be made bigger for the same electrical energy output.
Maintenance would be too expensive, so a space datacenter would have to be overprovisioned, e.g. with a double number of GPUs, to keep the average of the number of still working GPUs at the desired level over the lifetime of the datacenter. This will also increase the power and cooling requirements in the early part of a life cycle, and both the power generation and the cooling will have to be dimensioned for the maximum values, not for the average values.
It is certainly possible to design a datacenter that will work in space, but there is a very large number of difficulties that will increase the cost both in comparison with terrestrial datacenters and in comparison with existing satellites.
Until a detailed design is done by someone, it is not possible to know exactly how many times more expensive will be a space datacenter, but my opinion, as an engineer with experience in industrial design, is that the cost will be too high to make it worthwhile.
It's not a technical problem. It's a people problem. Every data center project over the last year has experienced some form of pushback no matter where they're built
By definition, a data center in space is remote, self-reliant on electrical power and doesn't use potable water. These features require technical solutions before they become practical, and once one has cracked those challenges, they have disarmed all the push-back arguments people have raised against data centers. It's telling that the data centers want to build-out near population centers, i.e. where someone else has paid for fiber, grid and water infrastructure; it's a good old fashioned raid of the commons. This is why no-one is pushing for building data centers in locales with 0.2 people/sq mile with no one to subsidize the costs.
Likely because the AI companies that are going to use those data centers have the absolute worst marketing imaginable. "We're going to take away your job and might accidently kill everybody on earth!" It's no wonder they get pushback.
> A remote town with a population of 43 will be happy for a data center to employ 2 locals, but data center operators aren't doing that today
This isn’t true, many of these DCs are in very very rural areas and get pushback from people not even living there. Governors get pressure from people hundreds of miles away of the site and then shut it down or impose moratoriums. It’s FUD on social media trying to slow AI advancement that is the root of all of it. The genie is out of the bottle, it’s like blacksmiths trying to slow industrialization. It’s already happened.
Not really. In sun synchronous orbit you can use solar power without battery backup. You can't do that on Earth because of night time.
Remote towns with populations of 43 are exactly the sorts of places that are pushing back on having datacenters in their areas.
And space is the ultimate physical protection. In reality a bigger issue than GPU thefts are protests or leftist groups trying to firebomb you. See the problems the Tesla factory in Germany has had with leftist violence, like setting its electricity pylons on fire (because electric cars "consume natural resources and are not sustainable", lol). The left has become very anti-AI and so the risk of attacks on facilities, especially if built outside the USA, is quite high which in turn means you have to overbuild physical security like high fences.
The number of slots in the sun synchronous orbits is limited. More slots can be accommodated if the satellites can maneuver to avoid collisions.
Moreover, a datacenter in a sun synchronous orbit must be able to maneuver anyway, otherwise without adjusting the orbit it will enter in the shadow of the Earth after a month or two, due to the revolution of the Earth around the Sun.
If the datacenter must make periodic adjustments of the orbit, it will need to carry a lot of fuel, especially because a datacenter will be much heavier than a typical satellite, so much more fuel will be needed for orbit changes.
Thus it is not certain that a sun synchronous orbit is the best choice. It is possible that the extra fuel required for it may make this choice worse than using another orbit where the datacenter will not have power for half of the time.
Antipathy towards AI isn't just on the left, you know. At least in the US it seems widespread, and the remote towns with population 43 are far more likely to have right wing inhabitants than left wing.
reply