The algorithm seems to have been failing to scale correctly, yes. On closer inspection, it looks like the billing software treated the bytes used as K used.

Did AWS use Grok to generate the billing sheets or the code for handling billing?

I'm serious. The errors reported look suspiciously like AI hallucinations or a signed integer being treated as unsigned.

Obviously billing errors are bad; but it seems like ones that off by an egregious number of powers of ten should concern us much less than ones that are small enough to be within the realm of plausible; those are the ones that you'll need to fight over and quite possibly not even win if you are an edge case or dealing with one of the services/configurations where you don't necessarily have any independent measure of usage. You can probably tell that you didn't use a VM more than 720 hours in the last 30 days; but are you actually counting GET requests in some way that is both authoritative and cheaper than just paying the $0.0004/thousand rather than hoping that Amazon will charge you correctly? For some very buttoned up buckets that only your other stuff accesses, quite possibly you can infer from those systems; but if it's something public facing and it might be a billing error or maybe you just got crawled hard last month?

Realbotix's classroom robot has drawn scrutiny because the company is connected to RealDoll, the longtime maker of hyperrealistic sex dolls and sex robots. Realbotix acquired RealDoll's parent company in 2024 but says the education-focused operation has separate employees, payroll, facilities, and technology, with plans to formally separate the businesses at the ownership level.

The "companion robots" are different from sex robots and intended to address what it's described as a "loneliness epidemic." Kiguel has previously said the company's goal is to produce robots and AI that are "indistinguishable from humans."

I'm not surprised surprised or anything; but it seems like a serious problem that it's the 'maker of high-end sex toys' part; rather than the 'attempting to replace education and human interaction with chatbots' part that has the company embroiled in controversy. Real Dolls are certainly pretty niche; a lot of additional inconvenience and cost for modest gains vs. vastly cheaper and more accessible local stimulation tools; but using tools as stimulation tools seems considerably less weird than using them as friends or fobbing children off on them.

I'd honestly be a little surprised if there's much protective effect.

It's possible that having to go through the setup process yourself is a little too much seeing how the sausage is made vs. interacting with a pleasant frontend cynically put together by someone who knows how 'engagement' works; but it's not like the nerd reputation for skewing socially awkward or the AI bro reputation for reaching a bit too quickly for slightly mystical anthropomorphisms are entirely unearned(the 'soul.md' is a frankly somewhat depressing genre).

At least for now; there might be some confounding demographic effects if you are talking one of the chunkier local models; in a country with a per-capita GDP of ~$14k being able to comfortably afford, or willing to uncomfortably afford, the necessary hardware would make you either at least modestly wealthier than average or significantly more interested than average; but as you slouch toward stuff that runs on more typical hardware the demographic differences presumably decrease.

Good question. Their POWER series of CPUs were not insignificant in capability, their chip designers were clearly technically sophisticated, and GPUs are just specialised vector processors with a few extra bells and whistles - stuff IBM is extremely familiar with.

It would not have been difficult to release a GPU or other LLM-specific processor to go along with the POWER11. They'd been working on the POWER11 for 4 years, they knew in 2020 that LLMs had a strong potential to be significant for Big Data processing - an area you use big iron for, they're not rank amateurs, they have plenty of reserve, they could have assembled an emergency team to build a vector processor that was custom-designed for just LLM work, and released an LLM processor card that could run circles around nVidia.

They didn't. Because, as has happened before, their management is simply too stupid and too slow.

I don't want to diminish the accomplishment; that seems like a very cool dataset and probably one that was really fiddly to pull together; but, if you are talking single-neuron resolution; I am curious about whether you can still call an individual sample "the human brainstem" rather than "a human brainstem" and what comparative purposes you can use it for without running into trouble with cases where there are multiple ways for a brainstem to be adequately healthy, so long as certain requirements are met, so you'll need considerably more samples to draw useful inferences about exactly what the problem abnormality is.

Same sort of thing as when "sequencing the human genome" was a big project. Obviously a major exercise in gene sequencing and a basis for situating subsequent sequencing operations; but once you start talking detail there isn't 'the human genome'; literally everyone has one; and it turns out that different differences matter or don't at radically different levels.

Presumably the methods used to do it once will be helpful in doing it more often in the future; but I'll be curious what we discover about the balance of 'normalcy' vs. some relatively subtle and confusing combination of surprisingly variable ways to have a brainstem that seems to work just fine along with surprisingly subtle, no ghastly big lesions, ways to have one that ends up being totally dodgy.

I find "NOTE: Experiences vary by region." to be a bad sign for something that would be so trivial for MS to alter the behavior of; and where they are obviously not earnestly making improvements that were previously impossible but grudgingly rolling back bullshit they thought they could get away with.

Probably means good news for users in the EU; same way they get left out of some of the most egregiously bullshit 'AI' stuff; may help EDU and enterprise; but I'm guessing that it's no promises for less favored users.

What if...

Someone (say someone who was familiar with doxygen and GCC) developed number of comment types, where some stipulated preconditions that must be true for the function to run correctly, postconditions that must be true once the function has run, kernel facilities that the function definitely needs, and kernel facilities that the function definitely doesn't need. These would all be optional for any given function.

A static checker could then validate if the code meets the behaviour expected by the programmer. This is precisely what is done in SPARK, a fork of Ada for high-reliability code. Combined with existing static checker capabilities, this would greatly increase the number of bugs that could be caught with all kinds of tools, AI included.

It could ALSO build a full fine-grained mapping for any fine-grained mandatory access controls system. You'd also want includes that you could import for precompiled libraries. This would allow someone to verify if the code was making unanticipated/undesirable calls but would also make SELinux possible to develop for at the application level.

It would not be trivial. If it was trivial, it would have been done simply because it already IS done in other languages and that makes it "obvious" to anyone who has been programming for a while. However, it should not be massively complicated, simply because you can use AI as the static checker. Once it has a definite set of bounda that must be satisfied, it should be much more capable of knowing what paths would violate those bounds. Which means that the checker stage essentially is trivial today, leaving only the markup stage.

I'm not expecting that from domestic opponents; both because the penalties are high and because people are, historically, shockingly bad at shooting for targets that actually matter. I'm thinking more internationally.

If 'AI' is half so interesting as its proponents claim one would expect being a machine learning researcher worth offering a fat signing bonus to be about as dangerous as being an Iranian nuclear physicist or a Russian oligarch who has fallen off Putin's friends list. If Zuck thinks that you are worth $100 million it seems like someone who takes the idea that 'AI' is the next frontier in state power would consider it worth the trouble to hire some local criminal to kill you in a botched robbery or have their clandestine services attempt to throw you a little tea party. So far no reports of even foiled attempts.

Anyone who thinks that the bots are going to wipe us out is overestimating the bots; but it seems like there is a disconcertingly large supply of people who are just chomping at the bit to replace people regardless of whether the bots are up to the job or not.

Honestly, the weirdest thing about the 'hard line activists' and the 'war with AI' is how much there isn't to it.

This is an industry that puts out a 'tehehe, we're an existential threat!' press release every time they need another VC round; and whose c-levels openly discuss how they will annihilate all human jobs and maybe someone should think about what we should do about that; and who routinely trample local interests to get their infrastructure builds rammed through; and what's the 'war with AI'? One idiot who tossed a molotov and a disgruntled constituent? That's it?

The same oddity is true for 'AI' companies and nation states, also very puzzlingly. To go by the rhetoric of 'AI' being an existential struggle for the future of industry and whether the AGI omnimind will speak english or mandarin you'd think that we'd see at least a bit of skullduggery. Prominent 'AI' hires occasionally dying under mysterious circumstances; sabotage of expensive GPU farms, maybe a Rosenberg-style show trial or two. But no. There's some lightweight hacking and ToS-violating 'distillation'; and a few export shenanigans; but aside from that it's basically the same as any other SaaS nonsense but with bigger numbers. Weirdly unserious.

There's also the issue of cost/scope. I'm not sure that Id is atypical for a studio of its age and size in terms of how its games have gotten more costly over time; but it's still very much the case that Quake was about a dozen guys crunching and Trent Reznor making creepy industrial noises; while Doom: The Dark Ages was closer to feature film level budget.

More potential players now than there were in 1996; but the production cost increase has still, on the balance, likely done bad things to your ability to turn a profit on the basis of a relatively small slice of the market thinking that your game fucking rules; rather than making a major seller more or less mandatory.

In Id's case specifically I'd also be curious if they are somewhat victims of their own legend. Basically every "boomer shooter" going is absolutely mainlining some combination of doom and quake nostalgia(often doom aesthetics but with quake's actually-3d to better render how you remember them rather than how the weird 2.5d stuff actually behaved); and a lot of those actually have dev teams closer in size to Quake; presumably with the expected effect on how many units they have to sell to remain viable.

Are we really just going to sit and take the fact that Cuba has already achieved the sort of electrical grid reliability that demented AI bros are setting billions of dollars on fire to try to bring to the US? This is an outrage!

"No, you don’t get to simply dismiss the oppression of Communism with a “I understand”. As if the death toll demanded by that ideology is dismissible with some shitty retort given on the ass-end of a customer support line."

Empirically, we do. You can feel however you want about it; but it's a trivial matter of historical record that, say, Vietnam, had trade relations pick up from the 1990s on; and full PNTR status in late 2006 leading in to WTO membership in 2007(both under noted woke liberal commie George W. Bush); and that's a nominally communist state directly continuous with the one we lost an unpleasant war to whose human rights record continues to be pretty tepid at best.

We do enjoy decrying the horrors of communism; but we'll 100% pick up an abjectly shit foreign policy pal if we think that they will be useful. During the cold war that normally meant any right wing dictatorship that liked us more than Moscow; now that Soviets aren't a huge deal you can still have "socialist republic" in your nation's full name so long as your resources are cool or your labor force is cheap and docile.

Cuba is really something of a weird outlier. Militarily unthreatening, not huge on strategic resources but some agricultural products we enjoy and by all accounts a pleasant vacation spot that (like a lot of pleasant vacation spots with dubious local governments) generally keeps things civil with tourists, even from unfriendly nations, so long as the tourists keep things apolitical and do tourist stuff. Normally that's the sort of place we'd absolutely do some business with.