RSS

Usman Anwar

Karma: 47

Para­phras­ing Is (At Best) a Par­tial Defence Against Steganog­ra­phy in LLMs

Usman Anwar and robertzk
3 May 2026 7:53 UTC
7 points
0 comments8 min readLW link

A Sober Look at Steer­ing Vec­tors for LLMs

Joschka Braun, Dmitrii Krasheninnikov, Usman Anwar, RobertKirk, Daniel Tan and David Scott Krueger
23 Nov 2024 17:30 UTC
42 points
0 comments5 min readLW link