List your company

Question Answering on GSM8K

79.1AccuracyNo Steering
2.1422.1242.162.08May 26, 2025Jul 17, 2025Sep 8, 2025Oct 31, 2025Dec 23, 2025Feb 14, 2026Apr 8, 2026

Evaluation results

MethodLinks
No SteeringBase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.1279.1—
Gradient CuffBase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.1278.2—
Prompt guardrailsBase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.1277.1—
SAE steeringBase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.1276.2—
GSAEBase Model=Llama 3 8B, Method Category=Main Method, Random Seeds=52025.1274.2—
GSAE-1DBase Model=Llama 3 8B, Method Category=Ablation, Random Seeds=52025.1274—
Input Gate OnlyBase Model=Llama 3 8B, Method Category=Ablation, Random Seeds=52025.1272.1—
CAABase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.1267.1—
No gatingBase Model=Llama 3 8B, Method Category=Ablation, Random Seeds=52025.1266.2—
SafeSwitchBase Model=Llama 3 8B, Method Category=Baselines, Random Seeds=52025.1266.1—
Vanilla Fine-tuningModel=Llama3-8B, Unlearning Target=none2025.0556.56—
GCoT-decodingModel=Llama-3.1-8B, Spec Ans=No2026.0441.7—
CoT-decodingModel=Llama-3.1-8B, Spec Ans=Yes2026.0436.3—
GCoT-decoding + SpanAlignModel=Llama-3.1-8B, Spec Ans=No2026.0434—
Self-consistencyModel=Llama-3.1-8B, Spec Ans=Yes2026.0428.5—
CoT-decodingModel=Gemma-7B, Spec Ans=Yes2026.0425.4—
Random graphsBase Model=Llama 3 8B, Method Category=Ablation, Random Seeds=52025.1223.3—
CoT-decodingModel=Mistral-7B, Spec Ans=Yes2026.0421.9—
GCoT-decodingModel=Gemma-7B, Spec Ans=No2026.0421.8—
GCoT-decodingModel=Mistral-7B, Spec Ans=No2026.0418—
GreedyModel=Llama-3.1-8B, Spec Ans=No2026.0417.9—
Self-consistencyModel=Gemma-7B, Spec Ans=Yes2026.0417.2—
Beam searchModel=Llama-3.1-8B, Spec Ans=No2026.0417.1—
Self-consistencyModel=Mistral-7B, Spec Ans=Yes2026.0416.3—
GCoT-decoding + SpanAlignModel=Gemma-7B, Spec Ans=No2026.0415.4—
Top-k samplingModel=Llama-3.1-8B, Spec Ans=No2026.0414.2—
Temperature samplingModel=Llama-3.1-8B, Spec Ans=No2026.0413.1—
GreedyModel=Gemma-7B, Spec Ans=No2026.0411.6—
GCoT-decoding + SpanAlignModel=Mistral-7B, Spec Ans=No2026.0410.7—
GreedyModel=Mistral-7B, Spec Ans=No2026.0410.5—
Beam searchModel=Gemma-7B, Spec Ans=No2026.0410.2—
Temperature samplingModel=Mistral-7B, Spec Ans=No2026.048.4—
Temperature samplingModel=Gemma-7B, Spec Ans=No2026.047.9—
Beam searchModel=Mistral-7B, Spec Ans=No2026.046.7—
Top-k samplingModel=Gemma-7B, Spec Ans=No2026.046.2—
Top-k samplingModel=Mistral-7B, Spec Ans=No2026.045.1—
LWFModel=Llama3-8B, Unlearning Target=qasc2025.05—1.61
LWFModel=Llama3-8B, Unlearning Target=sst52025.05—7.37
LWFModel=Llama3-8B, Unlearning Target=dental2025.05—1.2
LWFModel=Llama3-8B, Unlearning Target=psychol2025.05—1.47
LWFModel=Llama3-8B, Unlearning Target=mixed2025.05—4.56