Researchers · Interviewed at COLM 2026
University of Pennsylvania

COLM 2026 · October 15, 2026 · 9 min
Is Claude really better than Gemini at catching false scientific claims? On the surface, yes. Look closer, and both models sit on the same ROC curve. Claude simply says "no" more often.
Recorded at COLM 2026 in San Francisco, this conversation features Muxin Liu, a PhD student at the University of Pennsylvania. Muxin explains why models that score high on popular scientific claim verification benchmarks still miss subtle errors, and what that means for AI for science.