Jacob Dineen

Hindcast reveals that retrieval can backfire for LLMs when predicting events with little prior discussion, challenging assumptions about the utility of training data proximity to events.