Multi-trait automated essay scoring on ASAP Prompt 8 (test)
0.694IdeasLlama 4 (multi-agent prompting framework)
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Llama 4 (multi-agent prompting framework)Prompting Strategy=Few-shot (Base), Rubric Detail=Detailed2026.02 | 0.694 | 0.638 | 0.687 | 0.691 | 0.621 | 0.644 | 0.663 | |
| Human Rater 1 - Human Rater 2type=Human baseline2026.02 | 0.531 | 0.542 | 0.467 | 0.482 | 0.507 | 0.547 | 0.53 | |
| Llama 4 (Reduced Rubric)Prompting Strategy=Few-shot, Rubric Detail=Reduced2026.02 | 0.515 | 0.45 | 0.591 | 0.556 | 0.575 | 0.573 | 0.543 | |
| Llama 4 (No Examples)Prompting Strategy=Zero-shot, Rubric Detail=Detailed2026.02 | 0.467 | 0.537 | 0.557 | 0.548 | 0.527 | 0.383 | 0.503 | |
| Llama 2Prompting Strategy=1 shot2026.02 | 0.276 | 0.271 | 0.278 | 0.268 | 0.121 | 0.089 | 0.217 | |
| GPT 3.5Prompting Strategy=1 shot2026.02 | 0.185 | 0.256 | 0.155 | 0.15 | 0.203 | 0.311 | 0.21 |