Hallucinations vs. Predictions: Reframing Uncertainty in LLM-Generated Medical Responses
Track: Tiny paper track (up to 5 pages)
Abstract: Large Language Models (LLMs) are increasingly used in medicine, but the traditional factual/hallucinatory distinction fails to reflect the evolving nature of medical knowledge. This paper critiques that binary and proposes a refined, three-tiered classification: (1) Currently Verifiable Responses, (2) Tentatively Examinable Responses, and (3) Predictive Responses. This framework introduces a veridicality gradient and emphasizes temporal verifiability, enabling more accurate evaluation, reducing clinical risk, and supporting adaptive model calibration. Ultimately, it promotes the development of safer and more epistemically responsible medical AI systems.
Submission Number: 83
Loading