DAHL is a new biomedical benchmark and pipeline that breaks long-form LLM answers into atomic facts, checks each fact with a web-augmented model, and averages the true ones into a DAHL Score.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
DAHL: Domain-specific Automated Hallucination Evaluation of Long-Form Text through a Benchmark Dataset in Biomedicine
DAHL is a new biomedical benchmark and pipeline that breaks long-form LLM answers into atomic facts, checks each fact with a web-augmented model, and averages the true ones into a DAHL Score.