Task Semantic Diversity on RoboGene Generated Task Descriptions 1.0 (test)
80.32BLEU-1Rule-based
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Rule-based2026.02 | 80.32 | 67.82 | 62.11 | 56.45 | 57.86 | 69.84 | |
| Human2026.02 | 72.8 | 65.62 | 60.19 | 55.55 | 48.86 | 68.88 | |
| Gemini 2.5 Pro2026.02 | 38.97 | 9.76 | 5.73 | 3.23 | 27.67 | 34.63 | |
| GPT-4o2026.02 | 36.61 | 9.84 | 4.82 | 2.71 | 25.12 | 34.64 | |
| RoboGene2026.02 | 28.93 | 6.62 | 3.15 | 1.75 | 19.18 | 29.07 |