Ethereum's longtime Geth maintainer is flagging Anthropic's new Claude Fable 5 as an early example of companies deciding which knowledge domains stay open, noting the model's new safeguards shunt biology and cryptography queries to an older version and thereby blunt its everyday usefulness.
The scary part about Anthorpic's Fable nerf is not that it refuses to answer biology or cryptography. It's that it foreshadows what's coming. A world where a couple companies decide what you can and cannot do. They're building a new ruling class and you're not in it...
Fable 5 routes certain queries through extra classifiers that trigger on under 5 percent of sessions, deliberately conservative on biology, chemistry, and cryptography so the model can ship to a wide audience while the unrestricted Mythos 5 stays with vetted partners only.
Szilágyi and several high-profile engineers see the filters as a preview of concentrated control; Anthropic says it plans to tighten the biology and chemistry nets later, though no timeline or narrower criteria have been shared yet.
Many users criticized Anthropic for nerfing its AI model as an act of corporate censorship and paternalistic control rather than legitimate safety.
Yup, exactly this.
The scary part about Anthorpic's Fable nerf is not that it refuses to answer biology or cryptography. It's that it foreshadows what's coming. A world where a couple companies decide what you can and cannot do. They're building a new ruling class and you're not in it...
@peter_szilagyi Very precise observation.
In the capitalistic sense this is how they intend it to turn out,unlike Elon's direction of course.
But thank God for China.
China is doing all it can to undermine this,
With ever more powerful and cheaper models.
@aliomerhorzum @lex_node Well, they were also welcome not to scrape the internet and steal all copyrighted material - rather build up their own knowledgebase - but they ignore that, didn't they?
So your reasoning does not stand.
@peter_szilagyi You're sheltering unauthorized open weight models, are you not?
@peter_szilagyi @lex_node you are welcome to not use their product
it’s a for profit, and it’s entirely up to them to how much of their product they let their customers use
@aliomerhorzum @lex_node I think it's fair game to train AI on *everything*, if then you provide it as a public good. Fine if they don't want to provide the infra, as long as they release the weights.
But take everything and give nothing back is a bad place to be and will lead to a lot of animosity imo.
@aliomerhorzum @lex_node It was trained on copyrighted data, both from the internet and off internet. Their models were trained of theft. They don't get to claim ownership over it. That is my thesis.
@peter_szilagyi As opposed to the world before, where everyone had immediate access to experts whom to ask biology and cryptography questions, I guess 🙄
@peter_szilagyi @lex_node But those companies must obey government law rrright? And that government is democratically elected by people rrright?
@octo_bar The US designated cryptography as a weapon for a long time. Turned out that was stupid idea and using it everywhere is better, even with the occasional misuse.
@peter_szilagyi @lex_node literally every model (including open source models) are trained through data available on the internet
how does this validate your thesis?
anthropic models are not public goods
@peter_szilagyi This AI safetyism is getting really fucking annoying.
And most people just laugh and meme it.
@peter_szilagyi Elon won't let that happen. This is why @grok must win
@zengjiajun_eth @LefterisJP Whilst I'd agree and try, my project is about genomics and there are certain words I can't really avoid =)
@DellAnnaLuca The advantage of capitalism was that technology was diffused across the population, so whether you're the president or a lecturer, you can still use the exact same iPhone.
Once you start limiting access to technology, you're recreating a monarchy, and serfdom. That's the issue.
@peter_szilagyi @LefterisJP That’s why they’ll try to get open source models restricted next
@peter_szilagyi Sure. I’m just pointing out that the dark world you’re describing is half as dark compared to how dark the world was just a couple of years ago, at least for anyone except the top 0.1% or so for proximity to experts.
@DellAnnaLuca As I said, it's not about these particular two topics, it's about the trend and capabilities being built.
@peter_szilagyi When one company nerfs its own model, it hands competitors who don’t a direct opportunity to take market share. That’s the beauty of free-market capitalism. Open-source models are now only about a year behind frontier ones. It’s only a matter of time.
@CantillonFX @aliomerhorzum @lex_node So model distillation is fair game too, right? After all, my model just studies Claude and learns 😋
Ethereum's longtime Geth maintainer is flagging Anthropic's new Claude Fable 5 as an early example of companies deciding which knowledge domains stay open, noting the model's new safeguards shunt biology and cryptography queries to an older version and thereby blunt its everyday usefulness.
The scary part about Anthorpic's Fable nerf is not that it refuses to answer biology or cryptography. It's that it foreshadows what's coming. A world where a couple companies decide what you can and cannot do. They're building a new ruling class and you're not in it...
