Ah yes because the cost savings on solar power in constant sunlight over nuclear reactors or solar with batteries definitely justifies the cost of launching hundreds of thousands of tons in to orbit, including the miles of radiators that will be needed to cool all this. Oh and definitely justifies the cost of having to hire astronauts as technicians to repair the thing when something goes wrong.
The numbers for these data centers don’t even work on fucking earth, how does increasing the set up cost by an order of magnitude make this work?
Lmao, of course they backed off, they need to say they’re going to do it to get cash to keep going, but they can’t actually do it.
Because the moment they go for profit they’re gonna have to start explaining how they loose money on every single user of the product, all the way up to the highest tier of subscription. Then people might start asking hard questions like “wait…isn’t this more expensive to run than just having a person do these things?”
It’s so funny, because people act like open AI has a viable business model, but they’re loosing money even on their paying customers, even the highest tier of subscription. The product they’re selling really isn’t good enough to charge the price they would need to charge to pay for the operation costs, let along the training costs, and that’s with Microsoft giving them a bunch of servers for essentially free.
Like, there isn’t a path to profitability for them, certainly not on this scale. They’re just praying that if they throw enough data in to a big enough model that somehow it will start doing something different than what it currently does. It’s not a plan, it’s a prayer, a cult.
it goes deeper than just “investors are greedy” though. Most people making these investment decisions are doing it at the behest of other people who have handed them their saving in exchange for returns. Those people aren’t privy to the nature of how money is getting invested and why, they hire someone else for that, the investors.
The investors may be making short sighted, stupid decisions, but they’re doing it because they’re pursuing their own personal incentives, get a raise, a promotion, or just not get fired. The managers are doing the same. If they don’t do it, someone else will.
It’s not the fault or moral failing of any one individual, but a fault in the system of incentives. A failure in the fundamental structure of how we decide how investments are made, in how we accumulate capital for investment.
Perhaps there is a better term and I should be more clear, but people know, roughly speaking, what “new” does, even “active” is fairly straight forward. They are literally algorithms but not what people are talking about when they complain about “algorithms”.
When people complain about the “algorithm”, in the colloquial sense, they’re talking about some nebulous unknowable method of sorting that only the people at meta and alphabet are privy to the details of, not the literal definition of the word.
I should have chosen my words more carefully but I think the point stands, there is a marked difference between a system where it is clear to the user how things get sorted, and the home, discovery or “for you” systems of major social media sites.
Depending on how you browse, it was not algorithmically recommended. Even if you’re using “active” to filter, it’s barely an algorithm. Certainly not a personalized one, unless you’re just looking at the subscribed feed, in which case the personalization was done by you, not the formula.
That’s kind of the appeal of this kind of website, when there is automatic sorting it’s very straight forward and user mailable.
There is this interesting push and pull with algorithms, they need to show content users will engage with, but, their main value to the companies is that it allows them to easily manipulate what is seen.
They push people to hard they stop using the algorithm, but if they just let the algorithm act purely one what people engage with, then they can’t monetize it.
There is a third access of preventing people from going down self destructive rabbit holes, but they don’t care about that until people start talking about regulating them or start moving away.
Previously there was an obvious cap on the value proposition to scaling data centers, mainly, that they needed population centers nearby who would need storage or processing for thin film devices. Latency is important for these kinds of things, so they need to be near to the demands
Now they think they can make value regardless of demand from local population, through training weights for models, or running models and sending the output to population centers. So suddenly the cost of power to run the systems is what matters, and the most profitable (not the cheapest or most efficient) is fossil fuel.
They see dollar signs with the opportunity to turn power directly in to value without the need for people nearby.
It’ll be really embarrassing for them as the consumer market continues to fail to show interest in the outputs they’re making.
There definitely has been some scalping, but also, just, not a huge amount of inventory available (like sub 100 units available across cities with populations in the millions). A bit of a paper launch TBH.
TSMC only has so much throughput available and NVIDIA has other products they’re selling that they can make better margins on than consumer GPUs. I’m a little surprised they launched at all given how few they’re shipping.
I wonder how much of launching now was to generate buzz to get studios to adopt methods of rendering that work best with with software, make it harder for competitors to compete on hardware.
I can’t imagine they’d release a new chassis unless it was something radically different to their existing form factors, and even then, it would have to be a fairly big market sector, since they’re not really big enough to target anything niche.
Replacing an existing chassis would require that they continue developing and releasing new upgrades for the existing chassis in addition to the new one, or make all the internal parts interchangeable with one of the existing chassis, both options seems like an R&D nightmare for such a relatively small company. If they just dropped upgrading the existing chassis… well… that would kind of be counter to their ethos.
yah, that’s the point. They go for queries because each query, theoretically, means more ad views and more ad views means more money. So they made the search worse to increase query count.
Like, this was intentional and understood internally. Not saying the people who actually carried it out liked what they were doing, but management said to make number go up, so they did.