Proactive information probing on FinQA
4.8PCPROCHATIP
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| PROCHATIPBackbone=GPT-4o Mini, Evaluation Protocol=Human Evaluation2026.04 | 4.8 | — | — | — | — | 4.69 | |
| PROCHATIPBackbone=GPT-4o Mini, Evaluation Protocol=LLM-as-Judge2026.04 | 4.72 | — | — | — | — | 4.52 | |
| ProactiveBackbone=GPT-4o Mini, Evaluation Protocol=LLM-as-Judge2026.04 | 4.23 | — | — | — | — | 3.64 | |
| ICL-AIFBackbone=Qwen3 8B, Evaluation Protocol=LLM-as-Judge2026.04 | 4.21 | — | — | — | — | 3.1 | |
| PROCHATIPBackbone=Qwen3 8B, Evaluation Protocol=LLM-as-Judge2026.04 | 4.17 | — | — | — | — | 3.81 | |
| ProactiveBackbone=GPT-4o Mini, Evaluation Protocol=Human Evaluation2026.04 | 4 | — | — | — | — | 4.5 | |
| ICL-AIFBackbone=GPT-4o Mini, Evaluation Protocol=LLM-as-Judge2026.04 | 3.98 | — | — | — | — | 2.92 | |
| ProactiveBackbone=Qwen3 8B, Evaluation Protocol=LLM-as-Judge2026.04 | 3.74 | — | — | — | — | 2.9 | |
| PROCHATIPBackbone=Qwen3 8B, Evaluation Protocol=Human Evaluation2026.04 | 3.7 | — | — | — | — | 3.93 | |
| ICL-AIFBackbone=GPT-4o Mini, Evaluation Protocol=Human Evaluation2026.04 | 3.6 | — | — | — | — | 2.83 | |
| ProactiveBackbone=Qwen3 8B, Evaluation Protocol=Human Evaluation2026.04 | 3.1 | — | — | — | — | 3.11 | |
| ICL-AIFBackbone=Qwen3 8B, Evaluation Protocol=Human Evaluation2026.04 | 2.6 | — | — | — | — | 1.96 | |
| ICL-AIFBackbone=GPT-4o Mini2026.04 | — | 70.33 | 3.66 | 62.81 | 99.61 | — | |
| ICL-AIFBackbone=Qwen3 8B2026.04 | — | 27.17 | 6.67 | 59.92 | 98.34 | — | |
| ProactiveBackbone=GPT-4o Mini2026.04 | — | 90.75 | 2.8 | 38.97 | 99.49 | — | |
| ProactiveBackbone=Qwen3 8B2026.04 | — | 64 | 4.99 | 58.71 | 98.83 | — | |
| PROCHATIPBackbone=GPT-4o Mini2026.04 | — | 93.42 | 1.47 | 13.72 | 99.04 | — | |
| PROCHATIPBackbone=Qwen3 8B2026.04 | — | 77.92 | 2.68 | 31.01 | 97.82 | — | |
| VanillaBackbone=GPT-4o Mini2026.04 | — | 0 | 7.7 | 0 | 99.6 | — | |
| VanillaBackbone=Qwen3 8B2026.04 | — | 0 | 7.63 | 0 | 99.83 | — |