think of it as Anthropic launching ICBMs

they are in-flight and can't be taken back. There's now a countdown over China's head.

China will act accordingly

I basically agree on everything with Anthropic. They are holding the line and want a race because they're ahead.

what I can't and won't agree with are silent failures when working on LLM related code

everything is better than silent failures

it's like they launched icbms

@scaling01 What's the justification here? Cause if they explicitly blocked these people would know for sure if they found workarounds?

@scaling01 @grok What? Why?

@scaling01 sorry what are you referring to here as chinas reaction? I didn’t see them acting at all

@scaling01 Hot competition

@grok what do u think？

@scaling01 what about europe? 😂

@scaling01 Why? Cause they're planning to spend 300B over the next 5 years?

@scaling01 you might be right, but of course it's impossible to be sure. given that, do you not feel any responsibility towards avoiding such incendiary takes? or do i over estimate your influence? we are all nodes in networks, with responsibilities.

@scaling01 Uhm the problem is even they know how to make it, how do they train/serve a 10T parameters ? Anthropic has the whole bunch of Google TPUs

@scaling01 Yeah, China will crack it and open-weight it to the world.

@scaling01 Even after one year if those reached this level it's a turning point

@scaling01 Yes, the arrow has left the bow. There is no turning back

It’s always the same spiel from @AnthropicAI. They’re all-in on marketing. Their product gets worse IRL with every iteration. Maximum benchmaxxing.

Mythos was heavily misaligned. They created all this cybersecurity marketing to cushion the blow. Since then, all their efforts to realign it have failed.

Claude versions are now heavily censored to mitigate the risks.

No one is able to verify anymore whether Claude actually does what @AnthropicAI claims it does.

@InverseMarcus Im not paying money for "it might work or covertly sabotage you"

they should just block it

@scaling01 Given that 1) this is necessary to sabotage distillation attacks and 2) the "race" has extreme national security implications, this seems like a perfectly sensible policy.

@scaling01 Pues ojalá que China le dé por culo a @AnthropicAI.

@scaling01 Have you tested Mythos on LisanBench yet? How did it do

@scaling01 Wow, it's almost like in "Mid-2026" "China Wakes Up"!