Quick question: Do you want AI to be so well-trained that it can help husbands (or wives, for that matter) plan the perfect murder of their spouses? Probably not, right? Just as a gut reaction, it feels like a no. I wouldn’t even think it was a particularly difficult question.

But America contains many diverse perspectives, and one of those perspectives was shared by Comma AI founder and veteran jailbreaker George Hotz over the weekend.

The publication comes in response to a series of broad AI alignment plans, most recently the AI ​​Futures Institute’s AI 2040: Plan A policy document. That document envisions a world in which the world’s researchers collectively choose to slow AI development for 14 years for the good of humanity. But, of course, not everyone who reads the article agrees with its premises or conclusions.

Hotz is in the field and disagrees. In his post, he argues that the rapid takeoff scenario (the hypothetical scenario in which AI quickly gains superhuman abilities) doesn’t make much sense. I agree with a lot of what you say here! For Hotz, the best approach to AI alignment and security is to focus on locally controlled AI models that are closely aligned with the interests of their users.

It’s a good idea, especially because it reminds me how much of today’s AI relies on centrally managed services like Claude and ChatGPT. There are infrastructure-related reasons why AI services were developed this way: it’s expensive to host these large, cutting-edge models, and most people don’t use them enough during the day to justify truly personal AI. But those factors become less important as technology develops. Part of what was exciting about OpenClaw was this experimental, DIY approach, and it would be great to see more AI products try to bring that back.

But Hotz is a provocateur by nature, so he doesn’t stop there. Compare user-aligned AI to a gun (!), which doesn’t complain if you use it to kill your stepmother. (I feel like there are other rules against this?) A truly aligned AI could order meth lab equipment from Amazon Prime and show you how to use it if that’s what you want and ask for, he says. (Again, I don’t think AI is the limiting factor here.) Hotz even says that he would die to defend this principle, although it is difficult to imagine the series of events that would lead to that.

“Either we live in a world with freedom or we don’t,” Hotz writes. If those are the options, the world of freedom sounds better! Still, I don’t know.

It’s not all about freedom, right? Any structure that involves many people (societies, markets, corporations, etc.) requires balancing actions, linking individual needs in a network of preferences and interdependent systems of responsibility. And anyone implementing mass-market tech products should probably think about that network as a whole, which means taking seriously the interests of the world’s spouses and stepparents who haven’t yet been murdered.

The freedom that Hotz experiences is actually a space of potential futures made possible by collective enterprise; Those futures would disappear overnight if we all started behaving like little AI-powered Napoleons. As the meme says, we live in a society.

However, having a local AI willing to take on the corporate world for my benefit sounds great! I can’t wait for a review unit.