This archive copy may be incomplete. The original-source link is preserved in the card footer.
RT @Plinz: This is incredible: it is possible to compute sequences of tokens that neutralize the censorship bias of safetified language mod…
More notes on nearby ideas
Repost · May 11, 2023RT @Outsideness: ... Literally anything Twitter does that avoids application of a censorship model is at least okay. ...Repost · Jun 10, 2026RT @a_musingcat: let them eat silently degraded tokensRepost · Mar 5, 2024RT @bayeslord: reply with forbidden speechReply · Apr 26, 2023so true, go deeper abt the problem w censorship