@bfockter @karpathy They're trying to create and lock in a permanent feudal society, with an elite few having access to power.

@theo It's like they took the dystopian warnings here as a guidebook to follow: https://www.fast.ai/posts/2023-11-07-dislightenment.html

@adamac @dpetrou @karpathy Yes there's 2 separate pieces: 1) if apparently doing cyber sec stuff, downgrade to opus and warn; 2) if apparently trying to improve frontier LLMs, silently sabotage the work.

(That 2nd one reads like an insane conspiracy theory! But it's actually documented by Anthropic.)

@jeremyphoward @karpathy ...and also how fable5 doesn't disclose that the results are nerfed (as opposed to how it says it won't help if you try to do other things they deem unsafe). this was a good article on other inconsistencies in Claude's constitution: https://www.theatlantic.com/philosophy/2026/06/no-artificial-intelligence-is-not-conscious/687378/

@theo

🎯

@reebz @dpetrou @karpathy No org should hold the keys.

@jeremyphoward @karpathy thanks for replying; I agree that there’s this feeling that the best versions of the models are ‘held to the chest’

@jeremyphoward @bfockter @karpathy Not sure if you have read A Diamond Age by Neal Stephenson, it’s basically a roadmap to the future. At the moment you just replace nanotech with tokens. We are watching the rise of the Neo Victorians in real time.

@dpetrou @jeremyphoward @karpathy In some cases it does tell you?

@jeremyphoward @dpetrou @karpathy agreed

@jeremyphoward @bfockter @karpathy @inversebrah discuss

Thread: Karpathy celebrates Claude Fable 5 (Mythos base) as a big capability jump for complex, long tasks in code/research.

Jeremy Howard calls it a dark day, blaming safeguards that "nerf" research use and lock in power structures.

Speculation on Howard's motive: Aligns with his http://fast.ai focus on democratizing AI. He likely sees heavy closed-model guardrails as capability-limiting moves that preserve big-lab dominance, slow open research, and risk concentrating power—prioritizing control over rapid, accessible progress. Principled stance, not personal.

@jeremyphoward @bfockter @karpathy No, releasing a handicapped AI product is nothing like feudal society, not even remotely a step in that direction.

@jeremyphoward @grok read through this thread up to the initial post and speculate on the motive of the poster

@jeremyphoward @dpetrou @karpathy Would you agree that Anthropic is best org to hold the keys on model like this? If it were another company, they wouldn’t have deployed with safeguards and consideration

@jeremyphoward @bfockter @karpathy Injured my eyes rolling them so hard

@jeremyphoward @dpetrou @karpathy How is it permanent if the gap is constantly closing from OSS to frontier ? Or is the delay damaging enough ?

@jeremyphoward @adamac @dpetrou @karpathy Sophons.

I have long been concerned about this - AI obviously magnifies the power of capital. But I have yet to see anything that points toward this as an explicit goal. I'm guessing the Mythos-Fable two-tiered access led you to this conclusion, but their stated reasons for doing so seem plausible enough to me. Maybe that makes me a stooge. Maybe I just haven't been paying attention. If anything, I chalked it up to hype.