Fact Generation (Targeted) on WikiPeople
60AccuracyKREPE
Evaluation Results
| Method | Links | |
|---|---|---|
| KREPELLM=N/A2026.05 | 60 | |
| KREPELLM=N/A2026.05 | 59.1 | |
| Random FactsLLM=Gemini 3 Pro2026.05 | 39.6 | |
| Neighbor SetsLLM=Gemini 3 Pro2026.05 | 39.4 | |
| Iterative PredictionLLM=N/A2026.05 | 36.7 | |
| Neighbor SetsLLM=Gemini 3 Pro2026.05 | 36.1 | |
| Neighbor SetsLLM=GPT 5.22026.05 | 35.5 | |
| Random FactsLLM=GPT 5.22026.05 | 35.1 | |
| Random FactsLLM=GPT 5.22026.05 | 34.4 | |
| Iterative PredictionLLM=N/A2026.05 | 34.3 | |
| Random FactsLLM=Gemini 3 Pro2026.05 | 33.3 | |
| Neighbor SetsLLM=GPT 5.22026.05 | 30.9 | |
| Re-rankingLLM=Gemini 3 Pro2026.05 | 30.5 | |
| AutoregressiveLLM=GPT 5.22026.05 | 30.1 | |
| Re-rankingLLM=Gemini 3 Pro2026.05 | 29.3 | |
| Gibbs SamplingLLM=N/A2026.05 | 29.1 | |
| Gibbs SamplingLLM=N/A2026.05 | 27 | |
| Few-shot FactsLLM=Gemini 3 Pro2026.05 | 25.5 | |
| Few-shot FactsLLM=Gemini 3 Pro2026.05 | 25.2 | |
| AutoregressiveLLM=Gemini 3 Pro2026.05 | 23.9 | |
| Few-shot FactsLLM=GPT 5.22026.05 | 23.8 | |
| Few-shot FactsLLM=GPT 5.22026.05 | 23.2 | |
| AutoregressiveLLM=GPT 5.22026.05 | 21.3 | |
| Re-rankingLLM=GPT 5.22026.05 | 19.2 | |
| Re-rankingLLM=GPT 5.22026.05 | 17.6 | |
| AutoregressiveLLM=Gemini 3 Pro2026.05 | 17.5 |