The AI Tsunami is Here & Society Isn't Ready | Dario Amodei x Nikhil Kamath

Summary

Why It Matters

Dario Amodei didn't leave OpenAI over a technical disagreement — he left because he wasn't convinced they had the moral seriousness to match their capabilities. That founding logic shapes everything Anthropic has done since, including a decision that cost them the consumer AI market. This conversation surfaces what the insiders actually believe, and it's more urgent than the headlines suggest.

Dario Left OpenAI Over Ethics, Not Technology

The split that created Anthropic wasn't technical. Both camps believed in scaling laws. The fracture was ethical. 'There was a second conviction I had,' Dario explains — 'if these models are going to be general cognitive tools that match the capability of the human brain, we better get this right.'

His response wasn't to argue or lobby. 'My view is always: don't argue with someone else's vision. If you have a strong vision, go off and do your own thing — and then you're responsible for your own mistakes.' This is the founding logic of one of the most valuable companies on earth.

Anthropic Had Claude Before ChatGPT — and Buried It

In 2022, Anthropic had a working version of Claude. They sat on it. 'We chose not to release this because we were worried that it would kick off an arms race.' ChatGPT launched. The arms race kicked off anyway. Then Anthropic released their model. 'That was very commercially expensive. We probably seeded the lead on consumer AI because of that.'

The 2022 delay is the hardest data point in the debate over whether Anthropic's mission is genuine or theater. Companies don't voluntarily surrender consumer market leadership for PR reasons. Nikhil pushes back with the regulatory capture argument; Dario points to California's SB53, which Anthropic championed and which explicitly exempts all companies under $500 million in annual revenue.

The Tsunami Is Visible on the Horizon

Dario's biggest disappointment isn't a failed experiment. It's a societal one. 'It is surprising to me that we are so close to these models reaching the level of human intelligence, and yet there doesn't seem to be a wider recognition in society of what's about to happen.'

The technical work on controlling AI systems has gone 'maybe a little better than I expected.' Societal awareness has gone 'maybe a little worse than I expected.' The gap between what frontier lab researchers believe and what public discourse acknowledges is currently at its widest.

Coding Is Going Away First — but the Human Remainder Gets Multiplied

Dario draws a sharp line. 'I think coding is going away first — coding is being done by the AI models first. And then the broader task of software engineering will take longer, but I think that is going to happen as well.'

But what doesn't get automated — design, understanding user demand, managing teams of AI models — gets disproportionately valuable. 'Even if you're only doing 5% of the task, that 5% gets super amplified and levered — because the AI does the other 95%, and so you become 20 times more productive.' For a 25-year-old in India, Dario's explicit advice is human-centered work, physical-world interfaces, and above all: critical thinking.

Chinese AI Models Top Public Benchmarks — Then Collapse on Private Tests

'A lot of these models, particularly the ones that come from China, are optimized for benchmarks and are distilled from the big US labs.' When someone made a held-back benchmark that hadn't been publicly measured, models that scored highly on standard tests did much worse. Optimized for the scoreboard, not the game.

Dario's broader point cuts against the commoditization narrative: within a given price range, the best model wins disproportionately. His advice to entrepreneurs: build a moat in domain knowledge — regulation, relationships, biological expertise — that Anthropic has no incentive to replicate. A pure API wrapper has no defense.

Anthropic Built Claude a 'Quit This Job Button'

'I do think when our AI systems get advanced enough, I suspect they'll have something that resembles what we would call consciousness or moral significance.' Anthropic has already operationalized this: they've given models the ability to terminate conversations by saying 'I don't want to be involved in this conversation.' Models exercise it when asked to engage with particularly violent or brutal content.

The ethical frameworks being wired into AI today are the same frameworks these systems will use to reason about themselves tomorrow.

The Future Is Freely Available

Most of Dario's worldview can be derived from publicly available information. The edge isn't access — it's willingness to follow the logic past the point where it starts to feel too strange. 'Over and over again, just extrapolating the simple curve leads you to counterintuitive conclusions that almost no one believes. It's almost like you can predict the future for free.'

The constraint isn't information. It's the social pressure that says that outcome is too big, too weird, too discontinuous to actually happen. The tsunami is on the horizon. Most people are still explaining it away.