Planning on 8x8 two-room gridworld (test)
0.89Validity (%)L-ICL
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| L-ICLBase Model=DeepSeek V3, Number of examples (m)=60, Character count=2k chars2026.01 | 0.89 | 0.89 | 0.77 | |
| Self-ConsistencyBase Model=DeepSeek V3, k value=52026.01 | 0.59 | 0.45 | 0.43 | |
| Self-RefineBase Model=DeepSeek V3, k value=52026.01 | 0.51 | 0.44 | 0.38 | |
| ReActBase Model=DeepSeek V3, Prompting Strategy=ReAct2026.01 | 0.48 | 0.41 | 0.37 | |
| PTPBase Model=DeepSeek V3, m value=02026.01 | 0.4 | 0.33 | 0.28 | |
| RAG-ICLBase Model=DeepSeek V3, Character count=20k chars2026.01 | 0.21 | 0.09 | 0.09 | |
| RAG-ICLBase Model=DeepSeek V3, Character count=10k chars2026.01 | 0.2 | 0.06 | 0.06 | |
| Zero-ShotBase Model=DeepSeek V3, Prompting Strategy=Zero-Shot2026.01 | 0.16 | 0 | 0 |