Xenomirant

• Xenomirant 9 Aug 2026 1:25 UTC

  1 point

  0

  on: Don’t Inoc­u­late Every­thing: Strat­ified Inoc­u­la­tion Prompt­ing Nar­rows Back­doors and Pre­serves De­sired Traits

  Looks like a reasonable extension! Cool!

  Though, I wonder how is the oversampling approach different from a KL prior on the base model (the one without an inoculation prompt) for the mixed examples obtained after classification?

Build­ing Connections

Un­pack­ing Mul­ti­modal Data Leak­age, Bro­ken Bench­marks, and the Hes­sian Fallacy