Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

synthetic probes

Benchmarks

Task NameDataset NameSOTA ResultTrend
Candidate-ranking retrieval8192-token synthetic probes Weighted average
Accuracy50.6
4
Candidate-ranking retrieval8192-token synthetic probes Semantic paraphrase+distractor
Accuracy16.7
4
Candidate-ranking retrievalsynthetic probes Semantic paraphrase 8192-token
Accuracy50
4
Candidate-ranking retrievalSynthetic probes 8192-token Semantic exact
Accuracy52.1
4
Candidate-ranking retrieval8192-token synthetic probes Bridge-window
Accuracy59.9
4
Showing 5 of 5 rows