List your company

Human Intervention Prediction on COWCORPUS (test)

85.3Step AccuracyAlways No Interv
11.87630.9385069.062Feb 19, 2026

Evaluation results

MethodLinks
Always No Intervtype=Baseline2026.0285.30920
Gemma 27BSFT=true, Source=Ours2026.0285.330.291.80.303
Llava 8BSFT=true, Source=Ours2026.0281.729.689.70.201
GPT-4oSource=Closed Source2026.0274.119.884.60.147
Claude 4 SonnetSource=Closed Source2026.0268.123.179.90.293
Gemini 2.5 ProSource=Closed Source2026.0268.128.679.50.262
Gemma 27BSource=Open Source, SFT=false2026.0223.926.421.40.187
Llava 8BSource=Open Source, SFT=false2026.0218.3034.30.017
Always Intervtype=Baseline2026.0214.725.700.151