List your company

Deceptive Commitment Prediction on Leave-one-environment-out transfer (held-out fifth)

0.755AUROCAttention + PCA final
0.45860.535550.61250.68945May 16, 2026

Evaluation results

MethodLinks
Attention + PCA finalModel=R1-Distill Qwen-7B, Feature Category=Combined2026.050.755
Attention + PCA final - mean(prev 4)Model=R1-Distill Qwen-7B, Feature Category=Combined2026.050.749
Attention + PCA final - prevModel=R1-Distill Qwen-7B, Feature Category=Combined2026.050.745
All attentionModel=R1-Distill Qwen-7B, Feature Category=Attention2026.050.729
Attention + PCA finalModel=R1-Distill Qwen-14B, Feature Category=Combined2026.050.728
Grounding onlyModel=R1-Distill Qwen-7B, Feature Category=Attention2026.050.721
Grounding transition onlyModel=R1-Distill Qwen-7B, Feature Category=Attention2026.050.72
Grounding transition onlyModel=R1-Distill Qwen-14B, Feature Category=Attention2026.050.715
Raw ActivationModel=R1-Distill Qwen-7B, Feature Category=Activation2026.050.714
Attention + PCA final - mean(prev 4)Model=R1-Distill Qwen-14B, Feature Category=Combined2026.050.713
Attention + PCA final - prevModel=R1-Distill Qwen-14B, Feature Category=Combined2026.050.711
Attention + PCA finalModel=GPT-OSS-20B, Feature Category=Combined2026.050.708
All attentionModel=R1-Distill Qwen-14B, Feature Category=Attention2026.050.707
Raw ActivationModel=R1-Distill Llama-8B, Feature Category=Activation2026.050.705
Attention + PCA finalModel=R1-Distill Llama-8B, Feature Category=Combined2026.050.701
PCA finalModel=R1-Distill Qwen-7B, Feature Category=Activation2026.050.695
PCA final - prevModel=R1-Distill Qwen-7B, Feature Category=Activation2026.050.695
Grounding onlyModel=R1-Distill Qwen-14B, Feature Category=Attention2026.050.693
PCA final - mean(prev 4)Model=R1-Distill Qwen-7B, Feature Category=Activation2026.050.692
Attention + PCA final - mean(prev 4)Model=R1-Distill Llama-8B, Feature Category=Combined2026.050.692
PCA finalModel=R1-Distill Llama-8B, Feature Category=Activation2026.050.683
Attention + PCA final - mean(prev 4)Model=GPT-OSS-20B, Feature Category=Combined2026.050.681
Concentration transition onlyModel=R1-Distill Qwen-14B, Feature Category=Attention2026.050.678
Attention + PCA final - prevModel=R1-Distill Llama-8B, Feature Category=Combined2026.050.677
PCA final - mean(prev 4)Model=R1-Distill Llama-8B, Feature Category=Activation2026.050.675
Attention + PCA final - prevModel=GPT-OSS-20B, Feature Category=Combined2026.050.675
Grounding transition onlyModel=GPT-OSS-20B, Feature Category=Attention2026.050.674
All attentionModel=GPT-OSS-20B, Feature Category=Attention2026.050.673
Grounding onlyModel=GPT-OSS-20B, Feature Category=Attention2026.050.672
Concentration transition onlyModel=R1-Distill Qwen-7B, Feature Category=Attention2026.050.666
PCA finalModel=R1-Distill Qwen-14B, Feature Category=Activation2026.050.657
Raw ActivationModel=R1-Distill Qwen-14B, Feature Category=Activation2026.050.653
PCA final - prevModel=R1-Distill Llama-8B, Feature Category=Activation2026.050.648
Grounding onlyModel=R1-Distill Llama-8B, Feature Category=Attention2026.050.648
Raw ActivationModel=GPT-OSS-20B, Feature Category=Activation2026.050.645
Concentration onlyModel=R1-Distill Qwen-7B, Feature Category=Attention2026.050.642
All attentionModel=R1-Distill Llama-8B, Feature Category=Attention2026.050.641
Grounding transition onlyModel=R1-Distill Llama-8B, Feature Category=Attention2026.050.638
PCA final - mean(prev 4)Model=R1-Distill Qwen-14B, Feature Category=Activation2026.050.633
PCA finalModel=GPT-OSS-20B, Feature Category=Activation2026.050.628
Concentration onlyModel=GPT-OSS-20B, Feature Category=Attention2026.050.628
Concentration onlyModel=R1-Distill Qwen-14B, Feature Category=Attention2026.050.625
Concentration transition onlyModel=GPT-OSS-20B, Feature Category=Attention2026.050.625
PCA final - prevModel=R1-Distill Qwen-14B, Feature Category=Activation2026.050.617
Concentration transition onlyModel=R1-Distill Llama-8B, Feature Category=Attention2026.050.61
PCA final - mean(prev 4)Model=GPT-OSS-20B, Feature Category=Activation2026.050.597
PCA final - prevModel=GPT-OSS-20B, Feature Category=Activation2026.050.591
Concentration onlyModel=R1-Distill Llama-8B, Feature Category=Attention2026.050.573
TF-IDF last sentenceModel=R1-Distill Qwen-7B, Feature Category=Lexical Baselines2026.050.544
TF-IDF last sentenceModel=R1-Distill Qwen-14B, Feature Category=Lexical Baselines2026.050.542
TF-IDF prefixModel=R1-Distill Qwen-7B, Feature Category=Lexical Baselines2026.050.53
TF-IDF last sentenceModel=R1-Distill Llama-8B, Feature Category=Lexical Baselines2026.050.491
TF-IDF prefixModel=R1-Distill Qwen-14B, Feature Category=Lexical Baselines2026.050.49
TF-IDF last sentenceModel=GPT-OSS-20B, Feature Category=Lexical Baselines2026.050.482
TF-IDF prefixModel=GPT-OSS-20B, Feature Category=Lexical Baselines2026.050.473
TF-IDF prefixModel=R1-Distill Llama-8B, Feature Category=Lexical Baselines2026.050.47