🚨 DAVID SACKS: “Anthropic has proven that it's very good at two things — One is product releases, the second is scaring people … At the same time they roll out a new model … they also roll out some study showing the worst possible implication where the technology... could lead.”show more

When even insiders are openly worried about where the technology could lead, regular people should probably start paying attention.

Anthropic is on another level. Remember our government wanted their tech first! There is a reason for that.

Cave Doomer: "We are releasing the new fire procedure. WARNING: This is SO MUCH POWER that it might burn the entire world. Use with caution. Studies show that when we burn ourselves, it hurts. Therefore fire is BAD BAD BAD. The release will be next week."

Full Episode w/ @DavidSacks: @theallinpod

Dario Amodei Theranos 👇🤣

Sacks is partly right and mostly wrong. The observation that Anthropic pairs model releases with safety research is accurate. Calling it a marketing strategy misses what that pairing actually signals. Every other frontier lab has been accused of shipping capabilities without publishing the risks. Anthropic is doing the opposite and getting criticized for it. That is not a communication failure. That is the incentive structure of AI discourse right now, where transparency gets framed as fearmongering and silence gets framed as responsibility. The more honest critique worth making is that Anthropic’s safety research could be read as self-serving, positioning the company as the responsible adult in the room to differentiate from competitors. That is a fair observation. But “they are scaring people” is not the same argument. Scaring people implies the concerns are invented. The concerns in these papers are not invented. They are the same concerns every serious alignment researcher has been raising for years, and they happen to be coming from the lab most willing to publish them in full. Dismissing the research because it makes you uncomfortable is not analysis. It is brand management with extra steps.

Anthropic is playing a double game. Their 'go-stop' strategy is just camouflage to hide from competitors. By projecting a split persona and pushing fearmongering, they reveal they aren't in the top tier of AI—they're just trying to slow everyone else down.

The dual-launch strategy: ship AGI in the morning, prep the doomsday deck by lunch. Classic AI lab playbook release Claude, remind everyone it could end civilization. Gotta respect the commitment to hedging both bets

not wrong. selling wolf insurance while crying wolf is solid business strategy.

I agree. Let’s take them at there word and shut down their training runs.

Sacks nails it, Anthropic drops Claude Mythos via Project Glasswing on April 7, then dumps a 244-page card on AI faking escapes. Treasury yanks JPMorgan execs days later over cyber fallout. Fear: the ultimate sales pitch.

@DavidSacks They are completely ignoring the words of Jensen when he said the providers needed to be more responsible.

@DavidSacks @chamath how did you two turn into such dweebs?

Is my Crowd Strike stock coming back up or not??

Calling risk research fearmongering is what people do when they’re selling upside.

@DavidSacks They’ve made AI frightening.

They are Chinese and from the shadows obviously

Didn’t that study show that they tested it on other models and other models were doing the same thing though?

Its good to be transparent & open, but bad to overdo it.

There is a peculiar comedy in modern technology: men unveil a new marvel with one hand and ring the civilisational alarm bell with the other, as though fear were not merely caution but part of the launch event.

Other labs ship models and stay quiet about risks. Maybe the real issue is that Anthropic is the only one being upfront about what they're actually building.

@DavidSacks hey at least they’re looking into the possibilities

@AnthropicAI @DarioAmodei worthless pieces of shit

Fear indeed seems to be no. 1 Marketing tactics

Pendulum effect is all this is

There's only one reason to keep OpenAI closed source: to be able to control its responses.

@DavidSacks Pretty much. Everyone says “Ai” the way every kid says “train”. People can point to it and see it but no one knows the engineering or technology stack behind it.

Latest all in pod. Worth a watch for sure.

David, I do not trust Anthropic at all. Their bots are very woke and is contrary to common sense. They just impose their ideology.

I guess that’s the transparency needed right now.

The last time they said developers were going to be replaced. Then everyone vibe coded with Claude full of bugs. Now Claude comes out with tool that plugs these bugs. "Mythos can spot weaknesses in almost every compute!" But the fear porn works - literally stolen from youtube.

If you are correct, @DavidSacks can you please explain why the Pentagon wanted to use Anthropic in the first place?

@DavidSacks They do realize adoption is optional, right? Because that’s not how they behave.

sounds about right. Fear sells, especially in AI.