@emollick Not only have they not succeeded explaining it, but they have no response to what are obvious false positives, like nerfing the model for eg the Semi Analysis team and many others.

It sits poorly for paying customers when their vendor silently and deliberately nerfs the model

Why do they want to prevent people from doing things like distributed training in the first place? What is the rational explanation for that, other than attempting to prevent people from building open source models? And the whole claim about the false positive surface doesn’t ring true in the first place, it feels like an excuse.

@perrymetzger "I'm sorry, Dave. You can't do that." -- HALopic, 2026 A Safe Odyssey

@n00buntu I wrote that myself. I have no problem with using AI to write stuff, mind you, and don’t think it’s shameful to use AI.

@perrymetzger How can anyone trust they're not still secretly sandbagging? Once the mask is off...

@alexgrenier Apologetics, especially thin ones, are not going to change anyone’s mind. People like @DeryaTR_ can’t even ask an innocent thing like “what is a mitochondrion”, and claims that they can’t build a better classifier are horseshit, these things are amazing classifiers.

@perrymetzger It is quite obviously significantly better, because their previous actions were really bad.

@perrymetzger What they wrote was that the tradeoff to release Fable rapidly would be to have a large false-positive surface and that they're working to improve the classifier and reduce that as fast as possible.

@perrymetzger > That problem is not the specifics of this AI system, it is Anthropic and its ideological goals.

😆ifiuc you asked Claude to write a post criticizing Anthropic, just as the anti-data-centre activists used ChatGPT to write their diatribes?

Hilariously good fun.

@InquisitiveUrsa You can make something better and still not make it good.

@perrymetzger It's also impossible to know if they are even telling the truth. They could simply continue degrading performance silently at any time.

@perrymetzger Anthropic probably wants to wait until the heat dies down before doing the exact same thing again.

@perrymetzger Correct

@perrymetzger GPT has never refused any of my questions about AI algorithms in general nor to write code attempting to do them.

@perrymetzger Thanks for speaking up about this. Hopefully those with more influence on this platform continue to do so and keep some pressure on.

@perrymetzger Then we'll simply refuse to pay Anthropic. There are dozens of very good AI companies now, and not all of them are as pretentious and know-it-all as Anthropic.

@perrymetzger They're not ideological goals. That's just their PR strategy.

Analyze them like cynical economic agents.

@perrymetzger > I wrote that myself.

Ah. I thought perhaps you were poking fun at Anthropic.

And yes, everything you've written makes sense and illuminates the situation for me. Thank you. 🙏

Switching from silent sabotage to open refusal is a transparency improvement, sure. But it doesn't address who gets to define the boundaries, or why those definitions keep expanding to catch perfectly standard technical work. The policy change and the underlying ideology are two separate things, and conflating them does a lot of work for Anthropic here.