Web navigation on WebShop (50 instructions)
82.1ScoreExpert
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Experttype=Human2023.10 | 82.1 | 59.6 | |
| LATS (ReAct)k=30, n=52023.10 | 75.9 | 38 | |
| Fine-tuning2023.10 | 67.5 | 45 | |
| Reflexionk=302023.10 | 64.2 | 35 | |
| IL+RLtype=Reinforcement Learning2023.10 | 62.4 | 28.7 | |
| ILtype=Imitation Learning2023.10 | 59.9 | 29.1 | |
| ReAct (best of k)k=302023.10 | 59.1 | 32 | |
| ReActtrajectories=12023.10 | 53.8 | 28 |