Study: human and LLM everyday reasoning are both better explained as pattern matching than world models

arXiv·medium signal

Zach Studdiford and Gary Lupyan (arXiv 2606.13607, submitted June 11) tested human participants and 25 LLMs on common-sense causal reasoning and found shared, predictable error patterns triggered by irrelevant prompt details. They localize specific attention heads driving the pattern-matching behavior and argue human everyday causal reasoning is itself more consistent with pattern matching than abstract world models. The implication for agent builders: the gap between human and machine everyday reasoning may be narrower than assumed, but both inherit the same brittleness.