DeepMind Paper Argues LLMs Lack Abduction

A Google DeepMind paper titled LLMs can’t jump asserts that large language models handle induction and deduction yet cannot perform abduction. This gap prevents true hypothesis generation required for scientific discovery. Judea Pearl noted parallels to his ladder of causal reasoning, while Gary Marcus amplified related views that such systems may be a dead end for superintelligence. The analysis underscores fundamental limits in scaling current techniques without new mechanisms for causal abduction.