A new position paper published by Tom Zahavy, Research Scientist at Google DeepMind, examines a fundamental limitation of large language models: their difficulty with abductive reasoning, a capacity central to genuine scientific invention: LLMs can´t jump¹.

How do we fundamentally discover new things? In a letter to Maurice Solovine, Albert Einstein conceptualized discovery as a cyclical process involving an intuitive ’jump’ from sensory experience to axioms, followed by logical deduction. While Generative AI has mastered Induction (statistical pattern matching) and is rapidly conquering Deduction (formal proof), we argue it lacks the mechanism for Abduction—the generation of novel explanatory hypotheses. Using Einstein’s formulation of General Relativity as a computational case study, we demonstrate that the prevailing theory of “creativity as data compression” (induction) fails to account for discoveries where observational data is scarce. This position paper argues that while a modern Large Language Model could plausibly execute the deductive phase of proving theorems from established premises, it is structurally incapable of the abductive ’Jump’ required to formulate those premises. We identify the translation of simulation into formal axioms as the critical bottleneck in artificial scientific invention, and propose that physically consistent, multimodal world models offer the necessary sensory grounding to bridge this divide.

This (above) is Tom Zahavy’s reconstruction of the diagram Einstein drew in a letter to Maurice Solovine. It shows a cyclical line jumping from Sense Experience (E) to Axioms (A) via an intuitive jump (J), from which logical consequences are deduced.

And here (below) is the facsimile of Einstein’s letter dated May 7, 1952, addressed to Solovine—his close friend and the French translator of his works. (This excerpt corresponds to the second part of the letter, related to the “epistemological question” addressed in this post.)

Einstein was 73 years old:

We’re all doing quite well. My work capacity has decreased somewhat noticeably, though; well, that has its advantages.

Warmest greetings to you all!

____________________

(1) Zahavy, Tom. ‘Position: LLMs Can’t Jump’. Forty-Third International Conference on Machine Learning Position Paper Track, n.d. Accessed 2 August 2026. https://openreview.net/forum?id=klU4737opt.

Featured Image: AI (creative) reconstruction of Einstein’s diagram.