So, yes, they are cutting off search from AI-free schools. The claim is precise and accurate. This is the net effect, and they can, in fact, opt to not do this but they will not. There is no technical hurdle. There is no economic hurdle. Doing so simply does not align with their incentives.
There is no virtue in appealing to nuance to maintain plausible deniability for these companies. It's absurd on its face.
It is as much of a mistake to anthropomorphize billionaires as it is to anthropomorphize LLMs. People with resources greater than nations are .001% human and the rest is closer to a lovecraftian god. Consider that he can shatter markets or send massive trading companies under simply by having an affair. The empathy is misplaced.
Absolutely not. The value of empathy is not "have empathy because empathy is good", it is in the pro-social outcomes. Empathy is wasted, and counterproductively so, on the non-empathetic, especially on the powerful and destructive. It is a peace treaty, much like tolerance.
He jumped in front of that bus himself. If the investigation turns out he was definitely orchestrating defying export controls then it's more like he drove one bus into another. He deserves every bit of embarrassment and ridicule no matter how impressive his business accomplishments were up to now.
Are you and I looking at the same thing? Huang is a human, in every way you and I are. If he took a DNA test it would say he's human. Right now he is a human having what is probably the worst week of his life. A pretty good life? Sure. But empathy isn't about weighing what he "deserves" or making arbitrary judgements about how he should feel in regards to the level of wealth & power he wields. The fact is he's a human being who is experiencing a lot of pain. You should feel empathy for him and for the people he's hurt. If you don't, maybe you're more like an LLM.
> The model that underlies Claude Fable 5.2 and Claude Mythos 5.2 is a step forward in general capabilities. However, during the model’s training, we took steps to limit advances in the specific research capabilities that could most directly accelerate AI development. Thus, in relevant areas, including accelerator kernel optimization and improving large language model architectures, Fable 5.2 and Claude Mythos 5.2 perform at about the level of Opus 5.5. Importantly, we ensured that the training changes did not affect related capabilities like agentic coding, hardware development, and traditional AI or machine learning research (including interpretability research). We do not alter these capabilities as the model is being served to users (that is, at inference time). As described in Section 1.5 of the System Card and similarly to previous models, Fable 5.2 and Claude Mythos 5.2 also include transparent fallbacks to Opus 5 for a small set of topics in this domain, including kernel development on certain nonstandard accelerators.
-----
> Be it enacted by the Senate and House of Representatives of the United States of America in Congress assembled,
> Sec. 1. Every contract, combination in the form of trust or other- wise, or conspiracy, in restraint of trade or commerce among the several States, or with foreign nations, is hereby declared to be illegal. Every person who shall make any such contract or engage in any such combination or conspiracy, shall be deemed guilty of a misdemeanor, and, on conviction thereof, shall be punished by fine not exceeding five thousand dollars, or by imprisonment not exceeding one year, or by both said punishments, at the discretion of the court.
> Sec. 2. Every person who shall monopolize, or attempt to monopolize, or combine or conspire with any other person or persons, to monopolize any part of the trade or commerce among the several States, or with foreign nations, shall be deemed guilty of a misdemeanor, and, on conviction thereof; shall be punished by fine not exceeding five thousand dollars, or by imprisonment not exceeding one year, or by both said punishments, in the discretion of the court.
The implication of the comment is that Anthropic is limiting the capabilities of their external models to prevent other companies from building models that can compete with Claude, and that this constitutes anticompetitive behavior under the law.
Anthropic's "Pace the frontier" is a facially unlawful conspiracy to restrict markets. Imposing these restrictions in their product offerings is an overt act that they're taking to facilitate their unlawful market rigging conspiracy. (since their agreement with their conspirators won't have a full impact if upstarts or direct consumers can just take existing SOTA and use it to produce a competing product, routing around their oligopoly).
There is no legitimate purpose to restrict the ability to optimize accelerator kernels except to inhibit competition in this space-- it is the most benign and harmless act imaginable. There is at least some debate to be had about things like computer-security capabilities, but nothing is being damaged by AI inference being made 10% faster except the ability of market leaders to extract rents.
>...shall be punished by fine not exceeding five thousand dollars, or by imprisonment not exceeding one year, or by both said punishments, in the discretion of the court.
Wow, I believe violating Section 1201 of the DMCA carries a stiffer penalty these days.
That quote is from the original 1890 text. The statute has been amended and criminal violations are now felonies with bigger penalties including $100 million fines and 10 years in jail.
> punished by fine not exceeding five thousand dollars, or by imprisonment not exceeding one year, or by both said punishments, in the discretion of the court.
Something tells me that's not gonna do much to dissuade the AI cartels...
Real gamble, that. You're injecting something that no regulatory body that said was safe and that that supplier is giving you what they say they are? Just like injecting whatever it is they happen to give you?
If you're doing this sort of thing, usually you batch your orders together (ideally with some friends) and randomly lab test a few vials selected from the large batch you receive. The powder form vials you receive before mixing will last more or less indefinitely in a freezer, so you can order multiple years worth of supply at a time.
The potency and purity are typically pretty good but from time to time vendors will be outed for either outright fraud or a bad batch that is underfilled or low purity. Most of the direct vendors (out of China) will guarantee fill and purity, and also give credit on your next order for lab test costs if you share the tests with them for future marketing purposes.
All sorts of risks involved of course, but you can mitigate them fairly effectively. Your risk tolerance may vary. Some folks will buy single vials from US resellers and trust those seller's lab test reports. Others (smartly, imo) don't trust that and do their own testing directly on the product they receive.
This grey market supply chain is remarkably robust these days. Plenty of suburban Facebook soccer moms organizing group buys.
No, you can know that decisions are stupid before knowing their outcome. In fact, decisions can be stupid even when they result in a neutral or positive outcome.
the worst case for this is if that doesnt cause sufficient problems to force a reckoning. it has to hurt badly enough that we put significant effort into training legibility into the process.
the equilibrium state would be having found a good pattern for making patterns good. we probably aren't too far from that, so i think your worry may show up in the short term but not the long one
that would do it too, and i suppose people are working on that. but even without that, there arent infinite low hanging fruit problems to find, and any given code can just sort of be play-doh'd into other code. so the llms will need a few idk years worth of rounds of being trained on the things that pop up that wed never have the time to harden, and things will settle. like, its hard patching stuff up quick right now but we're managing and society hasnt fallen in to chaos.
What I have been reading makes me want to both agree and disagree with your take here. I think the fact that some of these things are not only established but are much simpler than one would think at first glance yields the much cooler realization that you can not only do a lot more than abliteration, but that humans can do it by hand if the thing is made legible.
From my experience with models when it comes to discernment about particular things, the models, even the best ones, still are not very good. The thing that we are most missing I think is acceleration in creating editing tools for humans that make the otherwise inscrutable both visually legible and interactive.
At least the Qwen3.5-0.8B loop bug isn’t that special (or rather not something worth extrapolating). I literally had this issue last week and told Astra to just go fix it.
Thats not what the issue is, it's that they aren't free to job-hop in the way that other people are. H1B keeps their wages artificially depressed and they have a sword of damocles over their heads threatening deportation.
There is no virtue in appealing to nuance to maintain plausible deniability for these companies. It's absurd on its face.
reply