cma

People don't always understand all the different theorems involved in a proof and can still gain from it. Can even do who layers with things like oracle machines where you just posit it finds a proof and opens up a whole rich hierarchy of stuff. Math understanding will probably get more like physics.

The caveat is they have to be a little more sophisticated than just changing rng on the sampling, because they have to detect for snippets in the middle of a long chat and with removed thinking. I don't know that any of the stuff they need for that actually has impact though.

> Another factor here is that criminal law in the US relevant to hacking requires intent. You don't want to go to prison for a software malfunction.

Do you think if you went out and purchased a humanoid robot and gave it a whittling knife and hooked it to an OpenAI LLM as the overall planner and set it out on the sidewalk in front of the OpenAI offices to make and sell crafted goods you wouldn't be charged if it killed someone, because you had no intent?

There is such a thing as criminal negligence, and you would be going to jail for a software malfunction, exempting all software malfunctions doesn't seem reasonable, some are worse than others.

Maybe hacking law doesn't have an option for that, but many people are just advocating change the laws when they say they should go to jail. A lot of people don't know you can't change the law to cover past stuff, but even ones that do may say something like that as a shorthand to pushing to change the law, not know the law doesn't already have an option (if it really doesn't), or just say a phrase to express frustration more strongly than saying something weaker.

Even if it is a user prompting the system, if a user asked for "make a fairly stable chlorinated molecule that can last long enough to spread through the atmosphere and induce contagious misfolding prions in people," I think if a hosted model that was near capable of it even started carrying out the first steps of research planning the whole company and its largest limited liability investors should go to jail, in all the senses mentioned above. Amend the constitution to allow ex post facto laws for that kind of stuff if models get close to that level of capability.

Weren't they planning to sell their compute? https://www.bloomberg.com/news/articles/2026-07-01/meta-is-b...

I guess at peak times they might need more to the extent they offload to azure and still sell excess off peak, or some announced partners required all data stay on azure?

See March 26, 2026 incident. Model not degraded, but harness changed to strip out past thinking tokens when a session went out of cache to save money and ease capacity constraints (affected API users too), resulting in bad degradation.

> you get severe slowdowns when weights are that large.

Not necessarily for MoE

But the 5090 they use there is now going low stock and selling for over $7500 in some places.

> What has no precedent is running it on an AI chat product.

Meta?

I'm not advocating for it, I don't know enough. But we already slaughter lots of animals and raise them in horrible conditions, and all robot assisted surgery already goes through animal trials.

> Second, this still doesn’t cover all the properties that make math proofs doable. It requires working in the physical world. You must physically capture or grow 10T cuts worth of animals.

It would likely be sim2real with that as the post training, reducing that requirement a lot.

> Cancer cells, for example, could regrow over months.

In that specific scenario you would likely train on receiving no unrelated injuries during the surgery, and have induced conditions with stuff tagged molecularly that you can then verify efficacy from without waiting months.

Depending on how much more data efficient sim2real makes it, you could end up seeing companies pushing it only for actual procedures the animals need but economically would never get; botched surgery and the animal gets euthanized before waking up, which they could argue was already going to happen.

> And most people wouldn’t trust the world’s greatest vet to operate on them, anyways.

Robotic surgery systems, already go through animal trials before being used on humans. So do many purely human surgery techniques.

Not all of it would translate, but robotic surgery could be done on anesthetized animals in a mass RLVR way.

Packages like BLAS

I was thinking more about this part

> Everything therefore becomes a hobby or a game. People do things because they enjoy them for their own sake, or because they are endeavours used as vehicles to socialise and enjoy others’ company, or because a shared social belief exists and is cultivated such that doing such and such a thing confers social status.

And the mother in the story:

> Vashti is content with her life, which, like most inhabitants of the world, she spends producing and endlessly discussing second-hand 'ideas'. She enjoys talking to friends but uses her work to defend herself against their invitations to be more social, remaining in her ‘room’ where all her basic needs are met.

She's basically something like an academic reduced to something more like wikipedia researcher while something else has the first-hand ideas, communicating with her circle through facetime/vision pro persona.

The Machine Stops by E. M. Forester is basically this world.

World locked rendering isn't a requirement for smartglasses. Some existing ones in glasses for factor do have more limited visual positioning systems. In more obtrusive formfactors they have positioning suitable for world locked.

Big money from taking Tesla's place in line for an Nvidia shipment during a huge shortage without compensation.

They already exist. Glasses used to have heavy (at high prescription numbers) glass elements instead of plastic. HUD glasses can already fall in that weight envelope, with better weight distribution.

In the xreal form factor it is pretty viable. It just needs a bit more resolution, and for more immersive gaming a bit more FOV, what it has FOV wise is fine for screen replacement.

> OpenAI themselves has admitted a weak version of this (that prompts might inadvertedly end up improving the model).

2023:

"The approach also aligned with the company’s broader deployment strategy, to gradually release technologies into the world for people to get used to them. Some executives, including Altman, started to parrot the same line: OpenAI needed to get the “data flywheel” going."

https://www.theatlantic.com/technology/archive/2023/11/sam-a... https://archive.is/NmO5P#selection-979.907-979.1177

I don't think this has been a big secret.

Xi is supposed to visit the White House soon and there are supposed to be AI talks.

They make giant ones too for bigger stuff. I think Adam Savage has a video about them for grouping unfinished project parts together.

I don't think it's as bad as that sounds; in math people work all the time with conjectures they aren't sure if true, and work out a lot of other interesting math based on whether it is or not. Something like Turing's Oracle machine gives lots of interesting math just assuming one could exist, even if it couldn't. It may be that there are things proved we can never come to a human understanding of, but still keep getting interesting math that relies on it that has aspects we can appreciate and enrich our knowledge from.

It's also possible he shared drafts of the work with someone else, who asked chatgpt to explain it to them with training on. Tao seemed to know lots of details of the work before anything was published, though also worked on the problem in the past with big results so maybe just guessed.

> and that humans are NOT making nice progress on

They've pretty much said their own work was heavily agent driven. Levent is in a particularly bad place here because while he probably had a lot of background in the Jacobian Conjecture problem, he made the solution to that one sound like someone asked the question and he just fed it to Fable during the world cup. Whether that nonchalantness was to just seem hip or was to promote Anthropic, which he has stock in, or was just the truth I don't know though. But it makes this one seem similar, when they might have had really had nearly a year of very valuable feedback to the models.

The vast majority of major global automakers program their vehicles to automatically unlock the doors immediately following a severe crash. I think they just weren't eliding the context and posting the full notes, not as criticism of this: nowhere in the commentary mentions it.

> Which seems to be referring to the unique ability to move the car, which other cars don't have.

Most cars with keys in the ignition/keyfob in the car after a crash have the ability to be moved if they are still operational (a requirement of this one too). I'm not sure about Waymo, the Jaguar one does still have a physical steering column though and isn't fully drive by wire.

For the base model?

I think Google does train on anything you put into docs if you aren't careful with the Gemini integration?

I've been asked over 15 times by Claude code to turn it back on (with the prompt defaulted on so if you accidentally hit enter your data is sucked in if you don't catch it).

I do think it ended up being after a power outage though and it was some kind of config corruption.

If you ask it not to think about something that doesn't cause the pink elephant issue?

If I heard Palantir-like I wouldn't assume something like a perfect mirror competitor company.

Interestingly of the two solved, both have rejected the prize money.