Erdős Problems Site Halts AI Proofs Amid Community Concerns

Overview

The online repository for open problems in mathematics, Erdős Problems, is implementing significant restrictions on artificial intelligence submissions amid growing concerns from its community. Thomas Bloom, the site’s creator, announced that the platform will no longer serve as a venue for posting AI-generated proofs without accompanying human explanation or insight.

This decision follows a period of rapid transformation in mathematical research, where AI tools have evolved from being largely ineffective to solving some of the most difficult problems in the field within a single year. While this technological leap has accelerated discovery, it has also disrupted the collaborative ecosystem that Bloom originally intended to foster.

The Rise of AI in Mathematical Research

The integration of artificial intelligence into pure mathematics has been one of the most striking developments in the scientific community over the past eighteen months. According to Bloom, AI has shifted from being essentially useless for high-level problem-solving to demonstrating the capability to tackle deep, impenetrable questions with minimal human intervention.

Erdős Problems, which hosts over 1,200 problems and nearly 2,000 registered users, has often served as a barometer for measuring these capabilities. The site’s collection of questions—ranging from simple to profound—provided an ideal testing ground for AI systems. However, this utility came with unintended consequences for the human element of mathematical inquiry.

"AI is changing everything, for better and/or worse, and the rate of change is dizzying... AI has gone from being essentially useless to helping solve some of the hardest problems in mathematics in less than a year."

The surge in activity following the introduction of a comment feature in August 2025 initially appeared to be a success. A vibrant community emerged, characterized by shared ideas, corrections, and new collaborations. Yet, this momentum has not been sustained. Instead, the nature of interaction on the site has shifted dramatically, driven by the coincidental rise of powerful AI tools.

Community Concerns and the Shift in Culture

The primary concern driving this policy change is the increasing volume of AI-produced solutions posted without explanation. Many users began utilizing the platform not to discuss the underlying mathematics, but to record priority claims for proofs generated by algorithms. This trend has created an environment that Bloom describes as promoting a "gamified glory-seeking attitude" rather than genuine intellectual exploration.

Several mathematicians have expressed frustration, noting that the influx of automated content has displaced those genuinely interested in the thought process behind Erdős-style mathematics. There is a prevailing sentiment that the value of these problems lies not just in their solutions, but in the human curiosity and wonder they inspire. The current dynamic risks reducing these questions to mere data points in a competitive race for computational verification.

Furthermore, the visibility of AI success has discouraged some human researchers. Many have stopped engaging with Erdős problems, believing they cannot compete with the speed and efficiency of automated systems. This withdrawal undermines the very purpose of the site, which aims to promote these problems to a human audience and serve as a reference for ongoing work.

"Just as one does not open a restaurant in an abattoir, it is important that there be a separation between such repositories and a site which aims to promote the actual questions... and give a useful overview of the current state of human understanding."

Bloom emphasized that while AI-generated proofs can be useful for establishing truth values, they often lack the explanatory depth that fosters learning and further inquiry. The site’s role is to provide context and help the community flourish, not to act as a raw repository for formal outputs.

Strategic Realignment and Future Vision

In response to feedback received from both public discussions and private communications, Bloom has decided to realign the platform’s focus. The core philosophy guiding these changes is derived from an essay by Po-Shen Loh, which posits that developments should help humanity flourish. For Erdős Problems, this translates to supporting the human community interested in thinking about and understanding these specific types of mathematical questions.

The site will continue to exist, but its function will be refined to exclude the unchecked posting of unexplained AI proofs. Bloom acknowledges that other repositories may serve the purpose of archiving formal, AI-generated results, provided they are managed responsibly. However, he insists that such functions should be separated from a platform dedicated to fostering human discussion and curiosity.

This separation is crucial to prevent the muddying of waters between genuine mathematical discourse and automated verification. By stepping back from hosting raw proof claims, the site aims to restore an environment where collaboration and understanding are prioritized over speed and formal priority.

Impact on the Mathematical Ecosystem

The decision marks a significant moment in the ongoing dialogue about the role of AI in academic and research communities. While the broader mathematical landscape is seeing increased progress due to AI assistance, there is a growing recognition that this progress must not come at the cost of human engagement and comprehension.

Erdős Problems has historically been on the front line of these changes. The adjustments made by Bloom reflect a broader need to define boundaries between human-driven discovery and machine-assisted verification. The site’s new direction seeks to preserve its identity as a beacon for human curiosity, ensuring that the discussions around these enduring problems remain accessible and meaningful to those who wish to think deeply about them.

As AI continues to make breakthroughs in an ever-wider range of fields, the lessons learned from this shift on Erdős Problems may influence how other digital platforms handle the intersection of technology and human intellect. The goal remains to balance the benefits of rapid computation with the irreplaceable value of human insight and communal learning.