Is AI solving open Erdős problems? (zeyu-zheng.github.io)

Recent developments in the use of AI for solving open Erdős problems have shown promising, yet cautious progress. Notably, a significant breakthrough was made regarding Erdős problem #728, where Kevin Barreto, with the help of ChatGPT-5.2, clarified the problem's ambiguous formulation and proved the adjusted conjecture using Lean. This indicates AI's ability to contribute meaningfully to mathematical solutions, demonstrating its potential to assist in generating original proofs, albeit with limited human input. However, many previous claims about AI independently solving Erdős conjectures were found to revolve around rediscovering existing results or clarifying ambiguous statements, rather than producing novel solutions. While AI has enhanced collaboration and efficiency in proving mathematical conjectures, most contributions have involved verification and formalization of human-supplied ideas. As we observe these trends, it is clear that while AI has not yet reached the stage of autonomously solving deep mathematical problems, its growing role in the field could eventually lead to significant advancements in tackling complex conjectures.