Fact Generation (Arbitrary Masking) on WikiPeople
64.9AccuracyKREPE
Evaluation Results
| Method | Links | |
|---|---|---|
| KREPELLM=N/A2026.05 | 64.9 | |
| KREPELLM=N/A2026.05 | 62.1 | |
| Neighbor SetsLLM=GPT 5.22026.05 | 49.7 | |
| Neighbor SetsLLM=Gemini 3 Pro2026.05 | 47.5 | |
| Random FactsLLM=Gemini 3 Pro2026.05 | 46.9 | |
| Random FactsLLM=GPT 5.22026.05 | 45.8 | |
| Iterative PredictionLLM=N/A2026.05 | 45.3 | |
| Random FactsLLM=Gemini 3 Pro2026.05 | 44.6 | |
| Random FactsLLM=GPT 5.22026.05 | 43.2 | |
| Gibbs SamplingLLM=N/A2026.05 | 41.8 | |
| Neighbor SetsLLM=Gemini 3 Pro2026.05 | 40.9 | |
| Gibbs SamplingLLM=N/A2026.05 | 39.7 | |
| Iterative PredictionLLM=N/A2026.05 | 39.5 | |
| AutoregressiveLLM=Gemini 3 Pro2026.05 | 36.4 | |
| AutoregressiveLLM=Gemini 3 Pro2026.05 | 35 | |
| Few-shot FactsLLM=GPT 5.22026.05 | 34 | |
| Neighbor SetsLLM=GPT 5.22026.05 | 33.9 | |
| Few-shot FactsLLM=Gemini 3 Pro2026.05 | 33.7 | |
| Few-shot FactsLLM=GPT 5.22026.05 | 32.4 | |
| Re-rankingLLM=Gemini 3 Pro2026.05 | 31.4 | |
| Re-rankingLLM=Gemini 3 Pro2026.05 | 31.3 | |
| Few-shot FactsLLM=Gemini 3 Pro2026.05 | 28 | |
| AutoregressiveLLM=GPT 5.22026.05 | 25.7 | |
| AutoregressiveLLM=GPT 5.22026.05 | 23.4 | |
| Re-rankingLLM=GPT 5.22026.05 | 17.7 | |
| Re-rankingLLM=GPT 5.22026.05 | 15.7 |