Jacob Dunefsky

One-shot steer­ing vec­tors cause emer­gent mis­al­ign­ment, too

Do safety-rele­vant LLM steer­ing vec­tors op­ti­mized on a sin­gle ex­am­ple gen­er­al­ize?

Transcoders en­able fine-grained in­ter­pretable cir­cuit anal­y­sis for lan­guage models

Case Stud­ies in Re­v­erse-Eng­ineer­ing Sparse Au­toen­coder Fea­tures by Us­ing MLP Linearization

Au­to­mat­i­cally find­ing fea­ture vec­tors in the OV cir­cuits of Trans­form­ers with­out us­ing probing