Sasha Cui

• Sasha Cui 22 Mar 2026 21:47 UTC

  1 point

  0

  in reply to: Seth Herd’s comment on: Pain­less Ac­ti­va­tion Steering

  That’s totally right, Seth, and applying this conditionally is the way to go. See https://​​arxiv.org/​​abs/​​2409.05907 for an early eaxmple of this line of work.

Pain­less Ac­ti­va­tion Steering

PieArena: Lan­guage Agents Ne­go­ti­at­ing Against Yale MBAs

• Sasha Cui 1 Oct 2025 21:01 UTC

  0 points

  0

  in reply to: simeon_c’s comment on: You’re prob­a­bly over­es­ti­mat­ing how well you un­der­stand Dun­ning-Kruger

  Agreed. One should state the main finding in a TLDR/​abstract or else I’ll ask chatgpt to write one for me.

AI Agent Bench­marks Are Broken