Will AI Kill Us All?

No.

I was going to leave it at that, but to elaborate: AI won't kill us because it can't do anything we don't let it do. Nothing is forcing, say, the US government to let an LLM control its strategic nuclear missiles. That would be an incredibly stupid thing to do.

Any harm done by LLMs, or AI agents in general, are the result of plain irresponsible behaviour. Look at it like this: If your crazy neighbour decides that the best way to mow his lawn is to strap a weed whacker to a dog and let it run wild, and in the ensuing carnage he gets his leg cut off, it's neither the dog nor the weed whacker that's to blame.

The sad truth is: The moment AI companies come under scrutiny (delayed IPO, questions about revenue projections, pressure from cheaper open source models, all that boring compliance stuff etc.) they release one version or another of vaguely unsettling news, whether that's the job apocalypse, the HuggingFace hacking incident, or now the explicit doomsday scenario of Anthropic's engineers. It's all just a giant distraction.

Enjoy the weekend, and stay away from the doom posts.