The AI Doom Publicity Stunt

Kevin Roose is mainly famous for suffering an early case of AI psychosis. He wrote an NY Times about how he was deluded into thinking that Chatgpt was in love with him.

Now he says in an interview:

"Hey, are we um are we all gonna die? Um, that seems bad. So, I [9:15] think this is breaking through and I don't want to be glib about it. The people inside the labs, they are genuinely concerned that the technology that they are building could kill everyone on Earth. That is not marketing hype. That is not a stunt. That is not just some clickbait that people are putting on their press releases. I have stalked to these people on the record, off the record, in public, in private. They say the same thing every time. In fact, they are watering down how likely it is that they think AI could kill us all in public. Their real projections are even scarier than what they'll say online.

He is still in a state of AI psychosis. If the AI researchers really thought that AI will kill us all, they would behave a lot differently.

Plutonium is dangerous, but with no potential to kill us all. The plutonium workers very carefully keep track of every ounce. The OpenAI and Anthropic workers do not take such precautions.

OpenAI brags about how it put an experimental AILLM in a sandbox, and it busted out. Nothing ever busts out of a properly constructed sandbox. If anyone at OpenAI really thought the AILLM was dangerous, then it would have been put in an air-gapped sandbox where it could not escape.

The three big AI companies are going public for a total amount that exceeds the total for all the tech IPOs of the last 45 years. Their talk of doom is to distract us from their real fear -- that free models that are almost as good will undermine their market.