Jiaxin Wen

Self-Recog­ni­tion Fine­tun­ing can Re­v­erse and Prevent Emer­gent Misalignment

• Jiaxin Wen 6 Mar 2026 18:51 UTC

  1 point

  0

  in reply to: micahcarroll’s comment on: Real­is­tic Eval­u­a­tions Will Not Prevent Eval­u­a­tion Awareness

  when the model is explicitly asked whether it thinks it’s in an eval. This probably leads to significant overestimation of eval awareness due to “reverse psychology” effects.

  yeah I think the classification eval tells us more about model’s capabilities than its propensity.

  A cleaner setup would be to sample both eval and production queries from similar distributions, collect the model’s responses, and then ask models to articulate the subtle differences between the two response distributions.

• Jiaxin Wen 28 Aug 2025 4:12 UTC

  2 points

  0

  in reply to: David Africa’s comment on: Aes­thetic Prefer­ences Can Cause Emer­gent Misalignment

  does anyone rerun openai’s persona feature tests on these new EM testbeds?

Un­su­per­vised Elic­i­ta­tion of Lan­guage Models

• Jiaxin Wen 4 Apr 2025 21:20 UTC

  1 point

  0

  in reply to: evhub’s comment on: Au­dit­ing lan­guage mod­els for hid­den objectives

  interesting! do you mean experiments in Sec 3.9.2?