I like Dwarkesh's style better than Lex Fridman, because unlike Fridman he's not a propagandist for Russia and doesn't have that "love" bullshit vibe.

But on the substance they're equally vapid. Dwarkesh's interview with Richard Sutton was especially cringe.

>What we’re learning from this episode is that the government actually has way more leverage over private companies than we realized.

Who is learning this for the first time only now? Even just restricting ourselves to the current administration, look at how many times Trump has directed punitive actions against private entities! Look at his actions against law firms like Perkins Coie or Covington & Burling. This is not something that just arose out of nowhere with Anthropic.

> But within 20 years, 99% of the workforce in the military, the government, and the private sector will be AIs.

I haven't seen this much hype and hopium since the dot com boom. The whole open AI -> Anthropic saga just reeks of the same evolution of Viant/Scient.

Look we have an amazing tool, but it has some fundamental shortcomings that the industry seems to want to burry its head in the sand about. The moment the hype dies and we get to engineering and practical implementations a lot is going to change. Does it have the potential to displace a lot of our current industry: why yes it does. Agents can force the web open (have you ever tried to get all your amazon purchase history?) can kill dark patterns (go cancel this service for me), and crush wedge services (how many things are shimmed into sales force that should really be stand alone apps). And the valuable engagement is going to be by PEOPLE, good UI, good user experiences are gonna be what sells (this will hit internet advertising hard for the middle men like google and Facebook).

I really hate how people think LLMs == AI. An LLM can’t/shouldn’t be doing anything other than generating text.

The uncomfortable truth is that "existing law already prohibits mass surveillance" has never been a meaningful constraint when the technical capability exists — we learned that definitively from the Snowden disclosures. Dwarkesh is right that the real question isn't whether current law is sufficient, but whether we want the only barrier between capability and deployment to be laws that intelligence agencies have historically treated as suggestions. Corporate resistance is fragile and inconsistent, but at least it introduces friction at the infrastructure layer rather than relying purely on policy compliance.

But the "Anthropic fight" is mostly fake. Palantir was using Claude as base model. Anthropic allegedly took issue with unsupervised kills because the technology wasn't ready (or something along the lines).

Also, I remember reading this guy has close ties to Anthropic. Also, I find it suspicious how he came to prominence out of nowhere. Like Big Tech and the establishment are propping podcasts of controlled narrative/opposition. I don't buy any of it.

Private AI's and searchable personal data troves are the only way to go if you care about privacy.

I speculate we'll discover there's very few unambiguously ethical uses of AI, much less for military applications. Them's the breaks.