Evidence Use and Identifier-Conditioned Prior Knowledge in Large Language Model Classification of Oncology Trials Assessed Through Progressive Content Removal and Counterfactual Testing: Comparative Analysis
{{output}}
Background: Large language models (LLMs) can accurately classify biomedical documents, but strong benchmark performance does not establish that predictions are grounded in the supplied text. In biomedical literature tasks, titles... ...
