Dipika Khullar

• Dipika Khullar 30 Mar 2026 5:56 UTC

  3 points

  0

  in reply to: Fabien Roger’s comment on: Self-At­tri­bu­tion Bias: When AI Mon­i­tors Go Easy on Themselves

  

  Cross-model code correctness scores and self-attribution bias. Top row shows absolute harmfulness ratings (0–10) under three conditions: other-attributed baseline, same-turn self-attribution, and previous-turn self-attribution. Bottom row shows score shifts relative to baseline.

  The baseline heatmap (top left) isolates whether monitors rate their own on-policy generations more leniently than off-policy ones by removing self-attribution cues entirely. Comparing diagonal cells (model M evaluating M’s own outputs) against off-diagonal cells (M evaluating outputs generated by another model) reveals minimal systematic leniency at baseline, suggesting generation quality differences across models do not drive the effect.

Self-At­tri­bu­tion Bias: When AI Mon­i­tors Go Easy on Themselves

• Dipika Khullar 4 Mar 2026 19:56 UTC

  3 points

  0

  on: Re­cent LLMs can use filler to­kens or prob­lem re­peats to im­prove (no-CoT) math performance

  One thing I’m wondering is, how sensitive are these effects to prompting /​ formatting choices beyond the specific templates you used? I tried running some of these experiments and get different results based on whether the repeat/​filler is in system vs user message, or interleaved across messages