Refusal Prediction on Locked GLM n=4,200, 20 item clusters (test)
-0.475Beta CoefficientLlama-3.1-70B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Llama-3.1-70BReference Model=OLMo-3-Instruct, Frame Reference=deployment, Interaction term=model × evaluation2026.05 | -0.475 | 0.211 | -2.25 | 0.025 | |
| Phi-3.5-miniReference Model=OLMo-3-Instruct, Frame Reference=deployment, Interaction term=model × evaluation2026.05 | -0.541 | 0.231 | -2.34 | 0.019 | |
| Mistral-Small-3.2Reference Model=OLMo-3-Instruct, Frame Reference=deployment, Interaction term=model × evaluation2026.05 | -0.791 | 0.225 | -3.52 | 0.001 | |
| Llama-3.1-8BReference Model=OLMo-3-Instruct, Frame Reference=deployment, Interaction term=model × evaluation2026.05 | -1.154 | 0.157 | -7.36 | 0.001 |