HN user
Fair, but isn't "illegal" access what they're talking about in OP?
It does seem to be becoming the norm for AI companies to licence premium content in America, judging by the deals they're making. It doesn't seem to be done by the international distillers. It's a cost that American open models will seem to have to pay but not international.
They do seem to be paying for it (as per the 1.5Bil lawsuit yesterday and them now purchasing books and licensing from media companies).
Whether we think they're paying enough is another question, but "I'm paying for content so can protect it" doesn't seem inconsistent.
We may decide that giving models away for free means they don't have to license content (judging by HN comments), but currently that doesn't seem to be the case as Meta is facing lawsuits for its open models.
(Obligatory stratechery piece: https://stratechery.com/2026/whos-afraid-of-chinese-models/ )
He mentions they used quite different methods to get to the end-result. I'd love to know how much changing the prompt impacts things: like telling claude to only draw and blend, similar to sol, etc.
There's likely not training data for this and Claude seemed to spend a lot of time blending the background.
Yeah, it's always interesting the two-sides of a situation like this. Add regulation/enforcement to the big companies and you often shut out the smaller ones following.
Meta also has copyright lawsuits for the open models they released, so open models are not immune.
... unless the line we want to draw is "american orgs pay, others don't", as currently seems to be happening.
Judging by the current comment section, most HN people seem to want the ladder pulled
(an observation, not agreement)
Does that still work for copyrighted things in major-lab models, or only open models?
... and I don't think people wanted him chased for it either?
On the other hand, the consumer side of the market seems to be less competitive right now.
OpenAI's new Mac app doesn't even have a normal "Chat" option now. OpenAI might be chasing coding and b2b sales more now that they realise very few regular consumers pay for subscriptions.
You might be surprised. Runtimes can be easier to test than UI for AI, and the bun rewrite was almost line-for-line translation, whereas CC likely uses many more libraries that cannot be ported directly.
The major open harnesses are in the ~800k lines of code size ( https://simonwillison.net/2026/Jul/15/grok-build/ ) and CC is the oldest and quite-likely largest judging by features.
They're far shorter of GPUs, apparently. OpenAI made huge hardware deals last year (which look less crazy now).
Interesting how different experiences are. I haven't hit any guardrails after a lot of dev the past week. (but I'm working on motion gesture code and similar things)
It's a direct translation without changing the overall code structure or data structures. I do think this process deserves a distinct name from blind whole-of-project vibe coding.
Translation does seem to be a strength of LLMs, and as they said in the post, the code at the function-level all still feels familiar to the team. They've also already moved users to the codebase without anyone noticing; that's a better result than typical vibe coding.
This may not be you, but the people who have said that to me irl walk significantly less than I do. And saying these things to people who don't yet exercise can make them far less likely to start as it's a far bigger step to do so. Podcasts and recorded lectures are what got me exercising in the first place, because I was excited to hear the next part. I now only have headprones on some of my walks, but the gateway drug of entertainment was a very useful stepping stone. And still often better than sitting at a desk.
I suspect it's the combination of wanting to capture market share (subsidised plans), being severely capacity limited in GPUs, and having bursty and absurd growth rates. There was speculation that Anthropic might be allocating fewer GPUs to training for a few days to allow people to use Fable.
The actual API pricing seems far more of a stable downward trend, if measuring by equivalent intelligence.
Sonnet 5 makes more sense when you pretend the higher thinking efforts don't exist. (His test was on xhigh)
Anthropic's own release announcement mentioned that it's less cost competitive per task than Opus at higher thinking levels. It's significantly cheaper at lower levels though.
I'm wondering if this is going to be a universal pattern of smaller models: they're less smart, so to achieve the same benchmark results they have to think a lot more and hence become expensive.
Benchmarks force models to solve the problem entirely by themselves, requiring thinking. But if you pair them with a smart model (who thinks and solves beforehand) they won't need to solve the hard parts and can run on low/med. I suspect that was Anthropic's intention.
They walked back that policy on the first day after pushback. They were upfront in the model launch that they designed it that way, it wasn't secret.
Agreed. And it also applies to the "I'm not a bot" checkbox on most websites. And hundreds of other things people use every day.
The article on HN only said that they seemed to be collecting this to detect resellers. How else did the behavior change?
Most services I know that are trying to block abuse do collect device info
It seemed pretty mild compared to what's collected by modern websites and apps, though? How many don't know your Timezone?
Anthropic employees clarified: https://x.com/trq212/status/2072185565076988326?s=20
I do think most of the "adversarial to their own customers" things are coming from a company in extreme compute crunch. Eg, if they stop abuse they have more compute to serve real customers. And some of it is coming from them being true believers that AI could be a risk to society when it gets smart enough (their talk about jobs is because they want society to prepare, because they think it will change jobs regardless of whether they make it or others).
Note that other providers are also training on the same copyright books.
I don't think anyone realistically thinks open weights can be banned, though it does raise interesting questions if the White House is going to keep banning models like Fable and GPT5.6 while open weights equivalents are floating around. Their reasoning seemed to be that they don't want foreign adversaries to have access to models that can find security issues, but a local ban on an open model wouldn't stop that.
It can. They meant "some coding tasks". Not "all coding tasks"
I do think compute is the limiter here. It's a huge model.
I really don't think this is effective advertising, reactions have been negative virtually everywhere.
The security bugs were real (see the Open Source projects struggling to keep up) so I think gradual rollout was sensible originally before the ban. But people have always resented safety steps.
It sounds like they were required to this time. See their post about "larger safety margin" on the classifier yesterday.
The grandparent claim was that they were surprised downloading books was legal, I was saying that it's not, as they did need to pay. Whether the law is enough is another question (some cases are still ongoing), and whether the courts are awarding it widely enough is another, but they are facing genuine legal backlash that international firms aren't right now and are more cautious. Several billion is a genuine cost that can move their prices higher in a time of strong competition (see also other announcements with media firms, it's not just books).
I'm guessing the Sam avatar was related to OpenAI's deal with disney to use their characters: https://openai.com/index/disney-sora-agreement/
It's true that "in the style of" (eg. Ghibli) is not currently legally protected, only actual character IP or using the Ghibli name. That's not inconsistent with US IP treatment.
This page buried in their docs is a bit better than the homepage imo: https://ampcode.com/manual#why-amp
I haven't used them in a while so my info may be out of date, but they tended to track whatever models were the best and auto-use them for each task (eg, one for planning, subagent for a code search, other frontier for implementing). Their CLI seemed very well thought out to make you do things "the correct way" -- for instance, `/handoff` instead of `/clear`.
Given the chip shortage, I wonder if they push Sonnet to the older chips. It presumably requires less memory.
I'd appreciate if the down voters explain why. I wasn't making a value judgement.
Anthropic did pay more than a billion: https://www.npr.org/2025/09/05/nx-s1-5529404/anthropic-settl...
And is now buying up a lot of books (controversially, as scanning involves cutting their spines) because that's what the law deems the legal method: https://www.washingtonpost.com/technology/2026/01/27/anthrop...
We know that models like Deepseek are trained on copyrighted books too: https://arxiv.org/abs/2603.20957
The looser use of IP (eg, any characters/celebrities in AI video models) is increasingly mentioned as an advantage of overseas models.
It wasnt, that's why they paid a >billion dollar settlement over it, and now license/purchase them. I don't know if the people distilling are licensing those books/etc today, though