Post

It is metaphysically impossible to build a safe general-purpose chatbot. Not difficult. Not expensive. Impossible. A system with an infinite goal space cannot be made safe. That is a structural contradiction. New post: www.jsnover.com/blog/2026/03...

30/3/26, 1:46 pm4 reposts12 likes

Hunnicuttt

@dmidlo.bsky.social

30 Mar

I disagree. It’s impossible to do it with coercion from the outside in.

Hunnicuttt

@dmidlo.bsky.social

30 Mar

Slight redaction: Yes, it is structurally impossible.

Hunnicuttt

@dmidlo.bsky.social

30 Mar

Down is up on these; up is still up. what if there were fundamental mathematics that we could salt the networks with? Same critique of the emergent intelligence work going on at the Pi group at Google. the fundamental assumption that the universe was formed from chaos. But what if 1+1 = Us.

Hunnicuttt

@dmidlo.bsky.social

30 Mar

Problematically, formally trained people are trained to train out the human heart in order to gain accurate and falsifiable information. But I don’t see that effect actually happening and higher order life on earth. Sterilizing a lab is safe for humans; bad for building intelligence.