Young Kwang Chae, Professor of Medicine at Northwestern University, shared on X:
“Proud to present our team’s latest research from Lurie Cancer Center at ESMO2026!
‘Characterizing Hallucinations in Large Language Models (LLMs) Using Real-World Oncology Cases for Clinical Decision Support’
Key takeaways from evaluating LLMs on 20 first-line NSCLC cases against NCCN Guidelines:
- High Accuracy ≠ Clinical Safety: Even top-performing models exhibit clinically significant hallucinations, emphasizing that high accuracy does not guarantee hallucination-free outputs.
- High Confidence Risks: Confident but incorrect responses pose critical risks to clinical decision-making, requiring strict human oversight.
- Input-Length Dynamics: The impact of input length on accuracy varies widely across models – longer inputs correlated positively in some models, but negatively in others.
Huge thanks to lead author Dr. Jiyon Lyu, and all co-authors for this impactful work on AI safety in oncology!”
You can also read:
25 Posts Not to Miss from ESMO Congress 2026: Day 4
