Score and Success Rate on Webshop
82.1ScoreHuman Expert
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Human ExpertType=Human baseline2022.10 | 82.1 | 59.6 | |
| ReActSetting=one-shot prompting2022.10 | 66.6 | 40 | |
| IL+RLTraining=10,587 training instructions2022.10 | 62.4 | 28.7 | |
| ActSetting=one-shot prompting2022.10 | 62.3 | 30.1 | |
| ILTraining=1,012 human annotated trajectories2022.10 | 59.9 | 29.1 |