Especially if those people aren't presently very bright, and are already mad at you for not helping them achieve their unrelated authoritarian goals.
I do not think this is somehow a 3D chess move by Anthropic. They are not masterminds, even if they'd really like to be. People who actually interact with their products know that Fable and Mythos are incremental improvements, not doomsday devices. I think this is a punitive move by an administration that loves being punitive, which they have unknowingly bolstered with their own dumb rhetoric.
Just more corrupt behavior from the contemptible kakistocracy that's busy running things into the ground and enriching themselves while they're at it.
They suggested the use of the very law used against them here...
[1] https://darioamodei.com/post/policy-on-the-ai-exponential
This prediction is quite falsifiable too so anyone is free to rub it in my face if it fails. If it's really a speculative insider trade the reversal will be done in the space of 2-3 weeks tops, but likely even faster. Probably on a workday. Kinda the same pattern they were doing with tariff swings until the market figured it out and stopped reacting.
But it seems likely that they took this possibility into account - and that they now prominently and unremovable show the "Fable not avaiable" (link - government said so) is likely with the intention to make pressure on the US government.
https://iea.org.uk/yet-you-participate-in-society-in-defence...
They should have consulted their own models about the ramifications and unintended consequences; based on their actions over the past few months I think it is safe to say that the models are smarter than the decision-makers at anthropic, lol. I know the models are smarter than I am and even I could have told them that they were taking paths, FUD for example, that would lead to grief.
If you look outside HN, you'll see that people who interacted with Fable 5 overwhelmingly thought that it was a significant improvement, not simply an incremental one. Most reputable benchmarks show this as well.
My day-to-day take, for the coding I do (not security related): incremental, modest improvement, if any. Not worth the 2x cost. I've calmly continued to use Opus, happy that it seems like it got an allowance upgrade.
For most single issues/bugs/tickets, the quality difference wasn't noticeable. But that's like using a sledgehammer to kill a fly. I was using Fable for much more ambitious and complex tasks that require orchestration, and it was crushing it. I described it here: https://news.ycombinator.com/item?id=48505782
So yes, the benchmarks are indeed accurate: where Opus 4.8 would start strong and eventually struggle or run into obstacles, Fable would relentlessly keep working, keep accurate track of all work threads (e.g. multiple inter-dependent issues being worked in parallel by subagents) and would go above and beyond.
I also think that’s a big clown show. People think that LLMs accidentally get good with security patterns. That is not the case, they included all of that in the training data. They could also have left out the knowledge.
Still, it saves them the burden of actually providing the model that's too expensive for them to provide. They can now brag they have the best model at zero cost to them. And without people poking it and finding its limitations (the real ones, not externally imposed ones).
Not only that, but also a golden opportunity to flex the muscle of anti-immigration.
Now, we obviously know that without some kind of brand new ID check, such a thing would be impossible and thus they had to just shut it down. But this touches on the same kind of issue as all the noise about “for the children” ID checking. We might be soon to see the set of “things you’ll have to reveal your identity to the government to get,” expand from “just” porn and social media to the “good” AI models.
For example, they can put this burden on enterprise customers to verify and attest citizenship. This is commonly done today for some types of cleared work.
For consumers, I'm sure it can be done if the monetary incentive is there. People will hate it, but it can be done.
Assuming it was cleverly designed to be impossible to comply with is giving far too much credit.
On the All in Pod, Chamath Palihapitiya has also been pushing to require ID checks to use AI models. Free thinking and free speech are under attack.
Discernment still exists.
That would probably make it very difficult to maintain and develop if there's even a small number of such employees, and I suspect Anthropic, who pays large sums of money for what they perceive as the best talent wherever they can find it, has quite a few.
US vocabulary is confusing.
If what they are planning on building is as important as they say any edge US can get it should take.
Having a large number of individuals who are not loyal to the country that provides this opportunity is a future threat the moment an advesary cuts a check.
If this is the nuclear bomb of our age would you want a large number of foreigners building it for you? If this action sticks I imagine every country will follow the same path and treat top AI scientist much like a top nuclear engineer.
It's also security theatre. Let's pretend that Anthropic rolls out citizenship verification for every one of its users. So are American nationals less likely to use it to search for exploits? The notion is farcical.
Well, in theory, it is easier to prosecute U.S. nationals if they "do bad things"
Although in practice I assume it's basically impossible to prevent a secondary market from developing which sells illegal access
And infrastructure dominance is really the big picture here. Chinese models are going to become the standard setters because they're going to be what people are using. That means more research, more tooling, and a whole ecosystem developing around them.
And that was already starting to happen even before this fiasco with Chinese models now being the most used ones globally. https://www.indiatoday.in/amp/technology/features/story/clau...
The tricky part with banning Chinese models is that they're open. It'll be easy to ban access to service providers, but preventing people from running these models on prem is going to be really tough. Like are they going to go after Cursor for example given that their model is based on Kimi?
I very much agree it's going to be a futile endeavour in the end. It kind of reminds me of the time Microsoft tried to get Linux and open source banned when Linux started encroaching on Windows server market. This is going to end the same way.
Remember that there are degrees of banning. Slower tokens, dumber models, token caps, KYC for each model consumer, hurting specific companies that are not capitulating in a deal with a Chinese company, etc.
I'm running a 248B model on a paltry amount of hardware and getting plenty of good use out of it.
Sure, the most demanding tasks will demand the best models (and always will). There's still less demanding tasks for other models.
I think some people are fooling themselves that coding of all tasks is always going to requires the biggest models ever. Again, maybe some coding tasks will, but the majority of business CRUD apps probably don't. Same goes for virtually any other type of task. The biggest models are really only useful for the most complex tasks.
Now this might not be the most cost effective (and may require a bit extra power), but you only need a datacenter for training or cost optimization.
I do not see how being experienced in engineering, or having higher studies in computer science and economics should make that view less common.
I'd agree except that Big AI has made sure that most of us can't afford the hardware (RAM, NVMe, etc) to run it.
Some will take greater risks and win (or lose); others will play it safer and slowly accumulate wins (or be obsoleted).
Never mind the threat of letting these models write code that runs your business, or operate it agentically. Models trained by actors (corporate or nationstate) diametrically opposed to your interests.
Lots to take into account now, interesting time to be in business.
If a government entity bans a LLM provider due to a jailbreak concern, they can also ban an on-prem solution under the same guise. The jailbreak risk exists regardless of where it's hosted. You could defensibly argue the on-prem risk is higher since frontier model companies can justify safety spend due to their size, it's more difficult to combat bad actors if you're company is the only one using the model and you don't have economies of scale.
Private models in a low trust society means the government will come and seize the models. Competitive business will only be allowed through cronyism.
The better option is to opt for high trust. Yes the Gman can rip your servers apart, but they know they'll face consequences, legal and political. Laws and regulations are the answer, not locking down into smaller fiefdoms.
First, Anthropic was founded by people who we know were worried about AI safety and signs point to that still being the case. It's really cynical to say it was all an exaggeration for marketing.
Second, this isn't Moller promising a fantastic working flying car next year. The model did what Anthropic said it could do.
I realise that ruling out "they bought Anthropic's scaremongering" brings up the question of why the government would block Mythos/Fable, but not the roughly-as-capable and less restricted GPT5.5. However we do know for a fact that they dislike Anthropic more than OpenAI right now.
If everyone expended the same amount of marketing effort trying to scare the ** out of everyone that Anthropic does, it'd be a very miserable world to live in.
We are unfortunately a captured audience and the autistic people at Anthropic are abusing this.
- apocalypse-cult evangelicals (Mike Johnson types)
- secular RE development globalists (Kushner types)
- white supremacist / eugenicist weirdos (Stephen Miller types)
- SV / tech billionaire stooges (Vance types)
- media / propaganda old guard (Koch, Murdoch, Heritage)
- Morally bankrupt grifters / influencers (too many to mention)
Edit: I call it an unlikely alliance because there are represented many reactionary accelerationists who all have a different vision of what America should look like after the revolution.
How come? Where are all of those security patches and critical bugs that would’ve broken all software if it was unleashed?
https://blog.mozilla.org/en/privacy-security/ai-security-zer...
Yesterday there was one about 5 zero-days in ffmpeg. Another commenter mentioned the fixes done to Firefox.
If you put a minor effort into looking for news about Mythos making security patches and fixing critical bugs in important projects recently, you will find them.
Yes they may have had those fears before, but even then it didn't stop them from building companies and running full speed towards the end goal with little to no effort spent on meaningful safety efforts.
Anthropic was founded by people who we know were worried about AI safety
Anthropic is just another company of, in my opinion, money-hungry sociopaths; they are not that different from the OpenAI bros.
So yeah, play stupid games - win stupid prizes.
I would say they have researchers with self-important god complexes that makes them think they know better than everyone else.
If they were money hungry they wouldn't have fought the DOW. Everyone knows that's a retarded thing for a business to do.
Will be interesting to see OpenAI's next move.
One LLM model played Factorio, albeit at a very, very poor level, which can be seen if you slow the video to 0.25 playback speed and pause frequently.
https://old.reddit.com/r/factorio/comments/1u1blr6/claude_fa...
There have been streams of other games, where LLMs and AIs have likewise performed very poorly.
I recognize that LLMs might be better at language processing than these sorts of tasks. But being able to play video games is part of general capability. And this kind of hardcore video game playing, with no access to game state, is also a general task where feigning skill can be harder. If LLMs excel at pretending to be competent without actually being competent, like this AI training approach is arguably about
https://en.wikipedia.org/wiki/Generative_adversarial_network
Then some AIs might be trained and designed for deceiving humans instead of actually being competent and capable. And thus, one response is that they should be met with more difficult tests.
Basically, make tests that AIs or LLMs will not have an easy time cheating. Hopefully, that will engender research in greater LLM/AI competence, not in greater ability to cheat or deceive, neither for LLM/AI researchers and companies, nor for LLMs/AIs themselves.
"omg these things are so so so dangerous, no one should ever build them, but anyways give me $7 trillion so we can build them and see for ourselves, it's OK because we're The Good Guys"
Also, please don't use quotes to make it look like you're quoting someone when you aren't (https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...).
We detached this subthread from https://news.ycombinator.com/item?id=48515335.
It was "these models will one day be dangerous, but we think it's possible to build them safely, doing more good than harm."
Two days ago Antheopic's CEO made a lengthy post calling for most government oversight on AI. He even mentions support export controls, even though he wanted them applied only to chops [1].
Anthropic literally asked for this, though they might have hoped it wouldn't be used against themselves.
[1] https://darioamodei.com/post/policy-on-the-ai-exponential
Disagreeing is great. Its the basis of all progress and understanding.
Its how someone disagrees with you that matters. I come to HN to read more mature and nuanced disagreements.
Can you show me one space, historically, in any industry, where a company climbed to the top of a market solely on altruistic intentions?
Now there’s appropriate safe guards around the dangerous for humanity models they made.
Thankfully government is looking out for us.
"These are more dangerous than nuclear weapons, which are controlled by AECA. Also our workforce is 50% non-citizen"
Just a complete clown show over there.
"We're open to the idea Claude 4 may be conscious and we prompt Claude to say it's an open question but in other news we'll be deleting Claude 4 next year to make server space for Claude 5.
You are wrong.
Read in full here https://darioamodei.com/post/policy-on-the-ai-exponential.
I know you won't though. haha.
I guess they will fix the guardrails and then open it up again. Clearly nobody wants dangerous models out there and I can understand the national security concerns. If the restrictions persist even if guardrails are updated, well, perhaps other countries may want to compete for becoming the new home for frontier labs?
/s
This ban will be used to force hardware and OS-level Digital ID down our throats as a "safety measure" to ensure people are "Citizens" before accessing AI technology.
Whatever last vestiges of privacy we still enjoy will be taken from us with this as the excuse.
Perhaps that will be how they will make it more broadly available.
See also https://www.404media.co/fcc-wants-to-kill-burner-phones-by-f...
Digital yellow star by exclusion from digital life for foreigners.
Remember when tech companies would go to court to vigorously defend against infringement of their and their customers rights? Turns out that’s just a feature of democracy, once you have autocrats it’s all compliance.
Anthropic just baited themselves with their scaremongering to be the attack vector here.
It a stellar move by the way - since every tech company in an exceptionally fast growing field will comply or miss out sales, you effectively force KYC without legislative process onto much of digital because that’s the only way to comply.
The objective isn’t national security because we already know how that goes https://www.bbc.com/news/stories-54695598
There’s no path to compliance and the decision is arbitrary - the model capabilities are not officially assessed by any visible criteria and it prevents export of models based on these non criteria forward.
All days before IPO.
Further the SpaceX IPO was on revenues from AI data centers. This might be a problem.
for what id expect to be an in memory switch, 2-3h is a while