I solved 6 open Erdős problems in 5 days, using OpenAI GPT-5.6 Sol (twitter.com)

In an impressive demonstration of AI's potential in solving complex mathematical challenges, a researcher successfully tackled six open Erdős problems in just five days using OpenAI's GPT-5.6 Sol. This achievement is significant for the AI and machine learning community as it showcases the efficacy of AI models in handling nuanced, high-level mathematical tasks that have stumped mathematicians, including prominent figures like Terence Tao. By strategically selecting problems that were already of interest to the mathematical community, the researcher managed to optimize the use of AI in this field. The approach involved crafting detailed prompts for Codex, an AI model capable of multi-stage reasoning and long-duration tasks. Each prompt was meticulously designed to guide the AI through not only the proof requirements but also potential pitfalls and adversarial checks on each proposed solution. This iterative method—consisting of attempts, diagnostics, and refinements—allowed Codex to navigate the complexities of the problems efficiently. The entire process, including proof PDFs, source files, and prompts, has been made publicly available on GitHub, indicating a promising step towards integrating AI into formalized mathematical research and potentially accelerating advancements in the field.