Fact Generation (Arbitrary Masking) on WD50K
60.4AccuracyRandom Facts
Evaluation results
| Method | Links | |
|---|---|---|
| Random FactsLLM=Gemini 3 Pro2026.05 | 60.4 | |
| KREPELLM=N/A2026.05 | 60.4 | |
| Neighbor SetsLLM=Gemini 3 Pro2026.05 | 48.8 | |
| Iterative PredictionLLM=N/A2026.05 | 44.7 | |
| AutoregressiveLLM=Gemini 3 Pro2026.05 | 42.3 | |
| Random FactsLLM=GPT 5.22026.05 | 40.5 | |
| Gibbs SamplingLLM=N/A2026.05 | 39 | |
| Neighbor SetsLLM=GPT 5.22026.05 | 35.1 | |
| Few-shot FactsLLM=GPT 5.22026.05 | 31.2 | |
| Few-shot FactsLLM=Gemini 3 Pro2026.05 | 31.1 | |
| Re-rankingLLM=Gemini 3 Pro2026.05 | 29.5 | |
| AutoregressiveLLM=GPT 5.22026.05 | 24.8 | |
| Re-rankingLLM=GPT 5.22026.05 | 13.3 |