Imagine you built a machine you could not fully explain, deployed it to a billion people in eighteen months, and then signed a letter comparing it to nuclear weapons. Then imagine you went back to work.
The letter and the Monday after
In March 2023, an open letter calling for a six-month pause on training AI systems more powerful than GPT-4 collected thirty thousand signatures — including the CEO of the company that built GPT-4. Two months later, a different letter appeared: one sentence, signed by the heads of OpenAI, Google DeepMind, and Anthropic. Mitigating the risk of extinction from AI should be a global priority alongside pandemics and nuclear war.
Then everyone went back to shipping.
That dissonance is the question. Not whether AI is dangerous — the builders themselves are telling you it might be. The question is what kind of danger, on what timeline, and whether the fear points at the right thing.
Four fears, one decade
The AI safety camp reads the capability curve and sees a species building something that will shortly be smarter than its builders, with no verified alignment method. Their fear is technical and clinical. The effective accelerationists are not ignoring the risk — they are arguing that the risk of slowing down is worse, that the real dystopia is the one where the technology exists and we refuse to use it. The governance-first camp spent fifteen years watching Congress fail to regulate social media and is now pre-grieving the same failure with a new noun swapped in. And the human dignity camp asks what it tells us that we built a system that talks before it thinks and handed it to six-year-olds.
Every camp agrees the technology is a discontinuity. Every camp agrees institutions are behind. The fault line is temporal. Safety researchers fear a future measured in years. Accelerationists fear a present measured in wasted months. The governance camp fears the gap. And the dignity camp fears something with no timeline — a slow forgetting of what it meant to think for yourself, to struggle with a question, to earn understanding the way you earn a scar.
Whether any of these fears is the right one may depend on a question none of the four can answer: does the thing we built understand anything, or does it just sound like it does?
The people building the most powerful AI on earth signed a letter comparing it to nuclear weapons, then went back to the office Monday and kept building. The argument isn’t whether it’s dangerous — they’re telling you it is. It’s what kind of danger, and on what clock.
Perspectives:
- AI safety
- Effective accelerationists
- Governance-first
- Human dignity concerns