Math and dot AI

Vaughan R. Pratt
Stanford University
Syntrees LLC

Topic 1: A "Severe Misalignment"

This page is in response to The Math-and-AI Declaration. For starters, I had this conversation with Microsoft's Copilot. I will add more material to this page as it comes to hand, and will also turn the page into a more interactive blog.

Topic 2: Feasibility of the "true airgap"

There are many situations where humans seek to align AI with their own values. But everyone's misalignment with AI is more or less severe. Humans like to comfort themselves with the thought that when the misalignment gets sufficiently severe, AI can be "airgapped" in order to contain the misalignment. This conversation explores the difficulty of creating a "true airgap".

My background

See here, here, etc. Put MATHAND in the subject line when writing to me on these topics at pratt@cs.stanford.edu.