LLMs Are Currently Not Helpful at All for Math Research: Hamkins (officechai.com)

Joel David Hamkins, a mathematician from the University of Notre Dame, voiced his stark critique of large language models (LLMs) in mathematical research during a recent Lex Fridman podcast. Contrary to claims of AI aiding in solving complex mathematical problems, Hamkins revealed that his interactions with current AI systems yielded disappointing results, often producing incorrect answers and failing to engage in constructive dialogue regarding its errors. His experience underscores a significant issue: the current state of AI lacks the reliability needed for serious mathematical inquiry, raising concerns about its practical application in research. Hamkins' skepticism is echoed by others in the mathematical community, such as Terence Tao, who points out that while AI can generate seemingly flawless mathematical proofs, they often contain subtle mistakes that are easily overlooked. This disparity emphasizes the critical challenge facing AI developers: bridging the gap between impressive AI performance in testing environments and its utility as a collaborator in real-world research contexts. As companies continue to invest in improving AI reasoning and problem-solving capabilities, Hamkins' insights serve as a sobering reminder that the journey toward creating effective AI partners in mathematics remains fraught with challenges.