Counterfactual sensitivity on OLMo-3-7B-Think Behavior Forecasting (test)
0.731Spearman CorrelationBehavior Forecaster
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Behavior Forecastertrained=true2026.06 | 0.731 | 0.931 | |
| Claude Opus 4.6reader_type=naive2026.06 | 0.69 | 0.888 | |
| GPT-5.4reader_type=naive2026.06 | 0.672 | 0.89 |