Decision Making on WebShop
42.9Success RateDual-Process (AUQ)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Dual-Process (AUQ)2026.01 | 42.9 | 0.888 | 0.782 | 0.755 | |
| Inverse UQ (UAR)2026.01 | 38.6 | 0.874 | 0.791 | 0.71 | |
| CoT-SC2026.01 | 37.1 | 0.861 | 0.769 | 0.663 | |
| RL w/ ACTBackbone=Qwen3-8B2026.03 | 33.8 | — | — | — | |
| Self-Reflection2026.01 | 33.4 | 0.855 | 0.735 | 0.615 | |
| IL w/ ACTBackbone=Qwen3-8B2026.03 | 31.6 | — | — | — | |
| Forward UQ (UAM)2026.01 | 31.4 | 0.865 | 0.775 | 0.725 | |
| Early Experience (Self-Reflection)Backbone=Qwen3-8B2026.03 | 31 | — | — | — | |
| Reflexion2026.01 | 30.7 | 0.84 | 0.742 | 0.63 | |
| RLBackbone=Qwen3-8B2026.03 | 29.4 | — | — | — | |
| ReAct2026.01 | 29.3 | 0.863 | 0.758 | 0.608 | |
| Imitation LearningBackbone=Qwen3-8B2026.03 | 28 | — | — | — | |
| ACTBackbone=Qwen3-8B2026.03 | 7.4 | — | — | — | |
| Prompt w/ CoT thinkingBackbone=Qwen3-8B, CoT=true2026.03 | 3 | — | — | — | |
| Prompt w/o CoT thinkingBackbone=Qwen3-8B, CoT=false2026.03 | 2.8 | — | — | — |