Theory: AI is a liberal democracy amplifier. The more authoritarian a regime, the more it will try to “align” AIs in clumsy ways that fail. 

AI breaks primitive ways of holding values such as explicit manifestos or even revealed preferences as in RLHF. Values-embodiment has to sink deeper into epistemology.

Blogger at ribbonfarm.com, running summerofprotocols.com, host of /gloom ☀️

Hopeful that this ends up being true. Maybe the best thing to come from the recent fiascos with Gemini + Meta's image generator is public awareness that a single monolithic AI system can't be aligned with an audience with diversity of values and a greater interest in a diversity of personalized AI systems.