It's an admirable idea, but I don't see how you can slow this down in practice. We certainly know China will not, and they are already at the frontier.
RSI is too important and too tangible at this point for everyone to slow down. It's the equivalent of telling the world to stop developing nuclear weapons after Hiroshima. Just not going to happen, and in the meantime dramatically increases the risk that the power of AGI lands in the hands of the few (read: the political class) rather than empowers the people.
I agree. I'm not even sure what the solution is (it's admittedly complicated given how people use ChatGPT today in so many different ways) but it certainly isn't this.
And this comes from someone that has most chats inside Codex, so I should be the least impacted by this!
I applaud the engineers that work at Anthropic, who have created both amazing products and uniquely intelligent models -- but I really shake my head at some of their business decisions and public comms which have done a lot to damage their trustworthiness in the business and developer community.
In just the past month: they decided to silently downgrade (instead of simply refusing) responses related to machine learning and other 'competitive' topics [1]. Then, they were caught fingerprinting certain request environments in a hidden way [2]. And now, once Fable is re-released after much frustration among its customers, they are providing it for a shorter period than promised (mostly over a major holiday period), with more stringent safety classifiers and a 50% haircut to usage limits.
It's hard to not view the organization as bizarrely adversarial to its customers. I was incredibly supportive of Anthropic during the supply chain debacle, as I viewed it as the capricious actions of a corrupt admin. But now I am wondering if it was just a response to the ineptness of their business leaders.
> It's hard to not view the organization as bizarrely adversarial to its customers. I was incredibly supportive of Anthropic during the supply chain debacle, as I viewed it as the capricious actions of a corrupt admin. But now I am wondering if it was just a response to the ineptness of their business leaders.
From the start Anthropic have been hostile to its own customers, and also trained on pirated books and had to settle north of $1.5B avoiding a $100B+ worth of damages if found liable.
Then they attempted and are still pursing against powerful open weight models by asking governments for regulatory changes that effectively ban the release of them - because it undermines their own moat (lol) and business model.
Now not only they were caught silently fingerprinting their customers requests, they are now placing ID verification for using their own powerful models, which could apply to everyone else for using powerful LLMs.
There just is no point in defending this company at all. Anthropic are NOT your friends.
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.
This is exceedingly easy to explain: demand is way too high, and the pro/max plans are loss leaders. I've paid a total of $20 and in 10 days, my cost, according to Claude code's cost tracker is like $400, which actually doesn't include all the use I've done.
Which is to say, if I continue my current usage over the month, I'll be getting $1000 of Claude for $20. It's difficult to be mad at someone selling me a $20 for two quarters, even if they're putting a bunch of restrictions on how and when I can do that.
I'm with you there. The way they treat their customers is high-handed and disdainful.
I'm gradually moving to GLM 5.2 on Opencode. It's the barest fraction of the price, and it's surprisingly capable. I notice very little difference vs. Opus 4.8.
There are awful models, and there are models nobody use, to paraphrase. Anthropic’s revenue skyrocketed earlier this year, according to their IPO filing. There has simply been too much demand. That’s the growing pain that everyone love to have, other than the affected users of course. That was why they paid a premium for all of the computing they could get from SpaceX, Amazon, Google.
I’m surprised to not see more commentary on this one. Much as folks may dislike AI and/or frontier labs, this is not good for capitalism, or democracy for that matter (given the actions of the executive govt today).
Seems the writing is on the wall for increasing inequality not just financially but now intelligence and economic opportunity as a result.
This will be particularly painful for startups and early stage businesses / SMBs that will be perpetually a step behind (likely multiple steps behind over time) companies with connections (especially those that are not above paying for connections in the admin).
I’d suspect bans on open source models to follow, and I wouldn’t be surprised if it hits hardware as well to fully close the loop.
> I’m surprised to not see more commentary on this one.
I agree -- this is big news, but the thread only has 21 comments?!
When the Trump admin kinda-sorta banned Fable a week ago, it seemed like it might be a one-off event: handicapping Anthropic because the administration has a grudge against them.
But today's news makes it seem like we're moving into a whole different world of AI regulation: each US model will have to be approved for release by regulators! And not only that, but the administration will whitelist who gets to use it "customer by customer." (Altman's words.)
> And not only that, but the administration will whitelist who gets to use it
This is the more dangerous part we should be terrified of tbh. Not being allowed to release it at all is one thing, whatever, just means we're capped at current capabilities for a while and things settle out.
The government picking and choosing who gets to access frontier intelligence is a huge issue and is creating the economic underclass all of us "skeptics" have been yelling at the clouds about since the beginning.
Found a startup? Well sucks to be you, your bigger competitors have access to more powerful intelligence than you do.
What happens when only the government has access to the powerful models, it will be wielded against citizens and non-citizens alike.
It also means we, the public, are no longer benefitting from the big infrastructure build out. The gains are going to be privatized, and yet agian, we bear the losses with no return.
It's enormous government overreach. A democratic government must not pick winners and losers. If they want to regulate the powerful models, it needs to be all or nothing. Either they cannot be released full stop or they must be available to the general public.
Startups and early stage businesses have always had less intelligence (when intelligence would be measured by the number and quality of their employees) than larger businesses. That hasn't stopped them from succeeding before.
I think there is a good chance that this is the AI lobby discovering a trick they can use to paper over lack of capacity by using their bought-and-paid-for influence on the Trump admin (not to be partisan: they've bought plenty of Democratic influence too, it just happens that Trump is currently holding the pen). Could OpenAI meet all demand for 5.6 if they wanted to? If not, wouldn't it be convenient if they weren't even allowed to offer it widely? Is this the same situation as Mythos/Fable?
AI and microchips should probably be treated like nuclear weapons and disease research. They all have profound non-military value, but powerful nations hoard them, build elaborate systems to deter proliferation, and reap most of the benefits. It's not exactly fair, but it's worked surprisingly well with some technologies over several generations.
But I don't see it happening soon, which probably means it will be too late. There's simply not enough competent political leadership in the world.
Surprised to not see more comments on this, especially given the popularity of the Anthropic/Karpathy article. What a win for OpenAI - and what a loss for Google, just 2 years after paying $2.7bn to bring Noam back into the fold. Does not bode well for Gemini long-term... Or could be a signal for how deeply they are leaning into world models.
Having worked in IB / PE my whole career (both buying & selling businesses and distressed debt, primarily in the middle market) it is always fun when my area of domain hits the front page and I can read the confident opinions folks have on my industry.
I think most people in this thread are missing the boat.
First, it’s important to realize that “default” does not imply wipeout. Default just means that you’ve violated a credit agreement, and that can be solved many different ways. Sometimes it’s solved by the sponsor restructuring the debt (injecting equity, asking for covenant/interest relief); sometimes it’s a true Ch 11 bankruptcy; in very rare cases, it can be a true Ch 7 liquidation. But even in those destruction-of-value scenarios, first-lien recoveries run 50-70% of “par value” borrowed. Lenders are smart, their job is to underwrite these deals, and they’re compensated for this with healthy interest rates - typically S+500-650 plus fees, which comes out to 10–12% unlevered returns. So they are basically pricing junk bonds, and in exchange get senior secured risk with real covenants around what the business can / can’t do.
Second, the myth that PE firms can just saddle a company with debt at the lender’s expense with no skin in the game and walk away with a profit while the company files for bankruptcy is hilarious, in part because it obviously doesn’t hold up to the slightest bit of scrutiny. The typical PE firm is buying a middle market business for 10-12x EBITDA on average. In today’s world, lenders will usually let you put 4.5-5.5x EBITDA on the business as leverage (in the ZIRP era, you could push things up closer to 7x, sometimes above that if it’s a stellar business). So on a typical 10x deal with 5x debt, 50% of the capital structure is PE firm equity which gets wiped out first before the lender loses a dollar. If that happens, the PE fund’s investor returns crater, which means no more Fund II/III/IV, which means no more fees for them to generate.
(As an aside, the typical “fees” paid by portfolio companies to the PE firm are, at least in modern limited partner agreements, largely offset against management fees and recaptured by the fund’s investors. The PE firm is not getting rich off these fees, at least not anymore.)
Third, private credit is not your local commercial bank. Many of the largest private credit firms are actually PE firms themselves - e.g., Apollo, Ares, and Blackstone, who are all known as “private equity” have actually become more valuable to the public investor community due to their private credit business. These are not sadsack regional banks and credit unions getting hoodwinked by New York finance elites. In many cases, they are the same firms, with the same resources (in-house restructuring / “workout” teams, portfolio ops, etc.).
It’s important to realize that private credit funds raise capital from institutions and HNW individuals with locked-up commitments - it is the exact same investor base as PE, and fundamentally a very similar business, they just invest at a different part of the capital stack. Because of this, risk of contagion is very low. Bank exposure to private credit is something like 1.5% of their portfolio - it is tiny. The extent of the blowback will be that a pension fund investing in alternatives has a poor return in one asset class across a dozen - it’s not systemic risk.
Finally, what IS true is that pre-ZIRP portfolios and software-heavy credits (something like 15-20% of leveraged loans outstanding) are in a tough spot. These businesses are either failing or they have too much debt that has “re-rated” to higher interest rates as interest rate hedges fell off. What happens from here is that the companies with structural issues in their business model will cease to exist (e.g., certain SaaS businesses). This happens all the time - it’s capitalism. But the good businesses are not going to disappear because of a bad cap structure - they will just get recapitalized.
It was fascinating to see OpenAI’s gaslighting in action last week. Signing their deal with the DoW and then announcing it so publicly clearly had the goal to (a) portray Anthropic as unreasonable actors that couldn’t come up with a “safe” solution like OpenAI and (b) take away all the leverage Anthropic had in the contract negotiations. Clever (in a Machiavellian sort of way) but still can’t understand why they did it so blatantly — literally hours after Anthropic was designated persona non grata by the government. Clearly this has backfired in a massive way.
In a way, I admire Dario’s stance and having the backbone to stand up to a government that is so happy to punish, legally or illegally, those that disagree with them. I certainly wouldn’t have the bravery (or stupidity) in his position — which frankly makes me happy that he’s running Anthropic and not someone like me…
