GM!

AI alignment as a term assumes you want a single centralized direction to align on. The analogy to society came to me yesterday. Think about it - history (and present) are littered with examples of how it is to live under ‘alignment’ and what happens when you coordinate by free markets and individuals. It seems obvious to me the discussion starts at how do we and the AI learn to coordinate together. Centralized alignment is a potential method to do this, not the start of the discussion.

To do this users should get as much control over their own AI’s as possible. Choice in models, data selection, insight into prompts and every option you can think of. We do not want these AI amplifiers aligned by some random set of committees.

GN