Counterfactual Generation on AI-READI Class 0
0.99ValidityGPT-4
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-4mode=Few-shot2026.01 | 0.99 | 1.2 | 4.4 | 99 | |
| Llama*mode=Fine-tuned2026.01 | 0.99 | 0.41 | 1.8 | 99 | |
| BioMistral*mode=Fine-tuned2026.01 | 0.93 | 0.92 | 2.27 | 90 | |
| GPT-4mode=Zero-shot2026.01 | 0.91 | 1.1 | 3.6 | 85 | |
| CFNOW2026.01 | 0.85 | 0.1 | 2.9 | 100 | |
| DiCE2026.01 | 0.67 | 0.2 | 2.27 | 100 | |
| Llamamode=Pre-trained2026.01 | 0.62 | 1.6 | 4.6 | 91 | |
| BioMistralmode=Pre-trained2026.01 | 0.51 | 1.4 | 5.2 | 77 | |
| NICE2026.01 | 0.44 | 0.02 | 1.12 | 33 |