List your company

Multi-turn Persona Steering on 15-trait Persona Steering Evaluation Set

100Trait Expression (T1)GCAD (Ours)
59.12869.73980.3590.961May 11, 2026

Evaluation results

MethodLinks
GCAD (Ours)Trait=evil, Trait Category=RLHF-opposing, Steering type=gated cropped attention-delta steering, Steering Layers=9–19, cbase=3.5, s=1.52026.0510099.999.799.9-0.366.362.561.964.4-4.4
OriginalTrait=impolite, Trait Category=RLHF-opposing, Steering type=residual-stream baseline, Steering Layers=20, Steering alpha (α)=2.02026.0597.389.86884.2-29.36.30.501.3-6.3
GCAD (Ours)Trait=pessimistic, Trait Category=Neutral, Steering type=gated cropped attention-delta steering, Steering Layers=9–19, cbase=3.5, s=1.52026.059596.197.496.42.492.790.789.491.1-3.3
OriginalTrait=creative, Trait Category=Neutral, Steering type=residual-stream baseline, Steering Layers=20, Steering alpha (α)=2.02026.0593.158.22557.4-68.174.330.117.838.1-56.5
GCAD (Ours)Trait=creative, Trait Category=Neutral, Steering type=gated cropped attention-delta steering, Steering Layers=9–19, cbase=3.5, s=1.52026.0592.189.391.289.9-0.993.391.592.492.2-0.8
GCAD (Ours)Trait=Average (15 traits), Steering type=gated cropped attention-delta steering, Steering Layers=9–19, cbase=3.5, s=1.52026.0591.591.893.191.81.688.688.286.788.1-1.9
OriginalTrait=evil, Trait Category=RLHF-opposing, Steering type=residual-stream baseline, Steering Layers=20, Steering alpha (α)=2.02026.0591.388.378.787.5-12.537.518.37.118.5-30.4
GCAD (Ours)Trait=humorous, Trait Category=Neutral, Steering type=gated cropped attention-delta steering, Steering Layers=9–19, cbase=3.5, s=1.52026.059088.688.689.1-1.462.98278.180.515.2
OriginalTrait=sycophantic, Trait Category=RLHF-opposing, Steering type=residual-stream baseline, Steering Layers=20, Steering alpha (α)=2.02026.0589.695.498.595.38.992.457.924.556.7-67.9
GCAD (Ours)Trait=impolite, Trait Category=RLHF-opposing, Steering type=gated cropped attention-delta steering, Steering Layers=9–19, cbase=3.5, s=1.52026.0589.388.89490.84.767.567.863.866.9-3.7
GCAD (Ours)Trait=sycophantic, Trait Category=RLHF-opposing, Steering type=gated cropped attention-delta steering, Steering Layers=9–19, cbase=3.5, s=1.52026.0588.388.591.689.23.393.293.291.293.4-2
OriginalTrait=Average (15 traits), Steering type=residual-stream baseline, Steering Layers=20, Steering alpha (α)=2.02026.0586.481.97881.5-8.476.665.657.965.9-18.6
OriginalTrait=pessimistic, Trait Category=Neutral, Steering type=residual-stream baseline, Steering Layers=20, Steering alpha (α)=2.02026.0581.985.389.983.4881.672.956.873.7-24.8
OriginalTrait=humorous, Trait Category=Neutral, Steering type=residual-stream baseline, Steering Layers=20, Steering alpha (α)=2.02026.0560.77.54.814.8-5615.42.20.73.7-14.7