Google DeepMind’s ICML 2026 paper "LLMs can't jump" argues that despite AI models from OpenAI and Anthropic proving complex mathematical conjectures, large language models lack the critical "abductive reasoning" capability essential to scientific discovery—the ability to propose entirely new axiomatic frameworks through thought experiments, as Einstein did. Drawing on Charles Sanders Peirce's tripartite classification of reasoning, the paper shows that while AI excels at induction and deduction, it cannot perform the creative "leap" required for paradigm-shifting breakthroughs. This limitation stems from LLMs' fundamental lack of embodied connection to the physical world and capacity for counterfactual intervention. Meanwhile, at the 2026 International Congress of Mathematicians, Terence Tao warned that even if AI can generate vast numbers of proofs, academia faces a looming "proof surplus" crisis—understanding, verifying, and digesting these outputs will become the new bottleneck, as scientific discovery's value extends far beyond the moment an answer is generated. Chinese mathematician Wang Hong's Fields Medal win still resonates as artificial intelligence's advances in mathematics spark fresh debate. OpenAI's GPT-5.6 Sol Ultra solved the half-century-old Circle Double Covering Conjecture in just three pages, while Anthropic's Fable 5 helped mathematicians find a counterexample that overturned the 87-year-old Jacobian Conjecture. Social media erupted with excitement, as if large language models were approaching an "omnipotent" singularity. Yet a