Fact Generation (Scratch) on WikiPeople
85.5AccuracyKREPE
Evaluation Results
| Method | Links | |
|---|---|---|
| KREPELLM=N/A2026.05 | 85.5 | |
| KREPELLM=N/A2026.05 | 77.7 | |
| Gibbs SamplingLLM=N/A2026.05 | 36.9 | |
| Random FactsLLM=Gemini 3 Pro2026.05 | 34.3 | |
| Few-shot FactsLLM=Gemini 3 Pro2026.05 | 32.6 | |
| Few-shot FactsLLM=Gemini 3 Pro2026.05 | 30.7 | |
| Random FactsLLM=Gemini 3 Pro2026.05 | 30.2 | |
| Random FactsLLM=GPT 5.22026.05 | 30.1 | |
| Re-rankingLLM=Gemini 3 Pro2026.05 | 28.8 | |
| Re-rankingLLM=Gemini 3 Pro2026.05 | 27.8 | |
| Iterative PredictionLLM=N/A2026.05 | 27.3 | |
| Few-shot FactsLLM=GPT 5.22026.05 | 26 | |
| AutoregressiveLLM=Gemini 3 Pro2026.05 | 23.8 | |
| Few-shot FactsLLM=GPT 5.22026.05 | 23.3 | |
| Re-rankingLLM=GPT 5.22026.05 | 23.3 | |
| Gibbs SamplingLLM=N/A2026.05 | 23 | |
| Re-rankingLLM=GPT 5.22026.05 | 22.9 | |
| AutoregressiveLLM=GPT 5.22026.05 | 22.8 | |
| Random FactsLLM=GPT 5.22026.05 | 21.2 | |
| Neighbor SetsLLM=Gemini 3 Pro2026.05 | 20.4 | |
| Iterative PredictionLLM=N/A2026.05 | 20.1 | |
| Neighbor SetsLLM=Gemini 3 Pro2026.05 | 19.8 | |
| AutoregressiveLLM=Gemini 3 Pro2026.05 | 15.4 | |
| AutoregressiveLLM=GPT 5.22026.05 | 14.5 | |
| Neighbor SetsLLM=GPT 5.22026.05 | 12.4 | |
| Neighbor SetsLLM=GPT 5.22026.05 | 10.9 |