LLMs Can't Jump

Scientific invention requires more than pattern recognition or formal proof. Following Einstein’s own account of discovery, it depends on an intuitive jump from experience to new axioms. Only after that jump does deduction and verification become possible.

Large language models have become strong at induction (extracting patterns from data) and are rapidly improving at deduction (reasoning from given premises). What they still lack is the capacity for the abductive jump itself: the generation of novel explanatory principles when existing ones are insufficient.

The critical bottleneck is the translation of simulation into formal axioms. Without a grounded way to move from an internal model of the world to new formal starting points, systems remain better at reorganizing what is already known than at inventing what is not.

Physically consistent, multimodal world models may offer a path across this gap by providing the kind of sensory and causal grounding that pure language models lack.

This Ledger Entry expands how the reader thinks by locating the limit of current AI not in scale or data, but in the specific cognitive move required for genuine scientific invention: the jump from experience to new principles.