Dario Amodei makes clear that slowing AI down does not mean Anthropic will stop releasing more advanced models. CBS: “Does that then mean that Anthropic is going to stop releasing more advanced models?” Dario Amodei: “It doesn't mean that. What it means is that we need to make sure... that every generation of models that we release is properly tested.” ---- From "CBS Sunday Morning" YouTube channel, (full video link in comment)show more

Next step: They demand for METR and other “safety” orgs to review open source models. Unsurprisingly open source models will never pass but the frontier models will. How convenient for Anthropic!

I am sure, they continued to find some alarming issue which are bothering them badly or is it a geopolitical thing?

So... Dario. You release models without adequate testing? Sounds like a YOU problem, dude.

Then do it and quit asking for intervention by a clueless bureaucracy

Rohan, you’ve been covering this extensively. Have you covered @DavidSacks’ response? (attached below) Perhaps I missed it, but I skimmed your profile and don’t see it:

@DavidSacks covered indeed, that 8mn view tweet :) David's is the strongest and the boldest voice in this situation. 🫡

Properly tested by whom, though? I don't move FaithGPT onto a new model until it's been through our own benchmark, and those scores are public so anyone can check them.

I wonder why everyone thinks China is a big problem in the AI race. It's really pretty stupid to assume that other countries will slow down AI advancement. I'll leave these here so people can judge for themselves

Okay, then make sure. Since when does the government run RL lol... this guy

Bro, this guy is lying, check his eye movements and body shuffle. He is scared

Yeah, i think we all need to pay close attention to the verb here. Slow <> Stop. And the vast majority of us don't know how to efficiently use Fable / Astra high as it is.

Poco le importa acabar con los empleos de la gente a don altruista

The ones who ship don't need to tell you their philosophy. The ones who talk are usually buying time before you notice they haven't shipped.

Yeah this is the bit that never added up for me, pace the frontier while still pushing the frontier.

Anthropic slows down. This means they'll drop 5 more models to hit $30B ARR before the burn rate kills them. Speed run or bust

cry me a river😭

It makes his call to pace the frontier a joke. Testing a product thoroughly before release is called responsibility and a basic expectation for any company - software or otherwise.

That means he wants the world to slow down development so that he can rule others with this.

please dear god let these damn tech bosses just stfu and build, I don't care about their manifestos, world plans, religions, or their cod-philosophies just build

And here we were imagining what not...😂😂😂

Dario knows that pausing model training unilaterally is economic suicide in a market where open weights catch up every quarter. The real shift is moving verification from post-training red-teaming into runtime safety sandboxes. As long as compute clusters cost $500M+ to train, no frontier lab can afford to sit idle while competitors continue pushing capability frontiers.

La prisa por "más avanzado" sin pausa para sentarse con el riesgo es exactamente el error que cuesta caro en cualquier sistema complejo. @rohanpaul_ai

This is an internal matter. Discuss with the investors, no need for a public brouhaha whatsoever.