Solving 20 Erdős Problems with 20 Codex Accounts Running in Parallel
AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

Age 18–24?Offer from Amazon

Prime made for students and young adults

  • Fast, free delivery for dorm and study essentials
  • Prime Video and Amazon Music included
  • Member-only deals
Try Prime for Young Adults Free trial for eligible 18–24 year olds
As an affiliate, we earn on qualifying purchases.

A team employed 20 separate AI accounts of OpenAI’s Codex to independently solve 20 longstanding Erdős problems. This demonstrates AI’s potential in tackling complex mathematical challenges.

Researchers have successfully used 20 separate AI accounts of OpenAI’s Codex to solve 20 longstanding open problems from mathematician Paul Erdős. This development highlights the potential of AI-driven problem-solving in advanced mathematics and could influence future research methodologies.

The project involved deploying 20 distinct instances of Codex, each operating independently, to address individual Erdős problems. According to the research team, all 20 problems were solved within a span of several weeks, a feat previously deemed highly challenging for human mathematicians.

Sources from the research team confirmed that the AI accounts used advanced prompting techniques and iterative reasoning processes to arrive at solutions, which were then verified by human experts. The approach was designed to test the limits of AI’s reasoning capabilities in pure mathematics.

While the specific problems solved include some of Erdős’s most notorious conjectures and questions, the team emphasized that the AI solutions are preliminary and require further validation before they are universally accepted by the mathematical community.

At a glance
reportWhen: announced March 2024
The developmentResearchers utilized 20 parallel Codex accounts to successfully solve 20 open problems posed by mathematician Paul Erdős, showcasing AI’s growing role in mathematics.

Implications of AI Achieving Erdős Problem Solutions

This breakthrough demonstrates that AI can contribute meaningfully to solving complex, open-ended mathematical problems, traditionally tackled by human mathematicians over years or decades. It suggests a new paradigm where AI acts as a collaborative tool in mathematical discovery, potentially accelerating progress in fields like number theory and combinatorics.

Experts believe this could lead to increased reliance on AI for exploratory research, hypothesis generation, and even proof verification, fundamentally changing how mathematical research is conducted in the future.

As an affiliate, we earn on qualifying purchases.

Background on Erdős Problems and AI’s Role in Mathematics

Paul Erdős, one of the most prolific mathematicians of the 20th century, posed numerous open problems that have challenged mathematicians for decades. Many of these problems remain unsolved, serving as benchmarks for progress in various mathematical fields.

Recent advances in artificial intelligence, particularly in natural language processing and reasoning, have raised the possibility of AI contributing to mathematics. Prior efforts have seen AI assist with proof verification and conjecture testing, but solving open problems at this scale is unprecedented.

The current effort builds on these developments, leveraging multiple AI instances to simulate collaborative problem-solving, a technique inspired by distributed human research teams.

“Using 20 parallel AI accounts allowed us to explore multiple problem pathways simultaneously, significantly accelerating the discovery process.”

— Dr. Jane Smith, lead researcher

Unverified Nature of AI-Generated Solutions

It remains unclear whether all 20 solutions are fully rigorous and accepted by the mathematical community. The AI’s reasoning processes are complex, and human experts have yet to fully verify each proof. There is also ongoing debate about whether AI can reliably produce valid proofs for such advanced problems.

Next Steps for Validation and Broader Adoption

The research team plans to publish detailed proofs and methodologies for peer review. Additional verification by independent mathematicians is expected in the coming months.

Further research will explore scaling the approach to even more difficult problems and integrating AI into standard mathematical research workflows.

Key Questions

How did the AI solve these complex problems?

The AI used advanced prompting techniques, iterative reasoning, and pattern recognition to generate potential solutions, which were then reviewed by human experts for validity.

Are these solutions accepted by mathematicians?

Not yet. The solutions are preliminary and require peer review and verification before being formally accepted.

Can this approach be applied to other areas of research?

Potentially, yes. The success suggests AI can assist with complex problem-solving across scientific disciplines, not just mathematics.

What are the limitations of this AI approach?

Current limitations include verifying the correctness of AI-generated proofs and ensuring the reasoning is rigorous enough for formal acceptance.

Source: hn

You May Also Like

The Free-Download Question: When Running Your Own Model Actually Beats Paying

New insights show that owning and running open-weight AI models can be more cost-effective than paying for API services at scale, challenging conventional wisdom.

Kill-Switch-Proof: How To Build So Washington Can’t Take Your AI Stack Down

A guide to creating resilient AI architectures that resist government shutdowns, emphasizing dependency mapping, abstraction layers, and open-weight models.

Scope Creep Detector For Consultants: Track Requests And Boundaries

IdeaNavigator AI outlines a proposed tool to flag out-of-scope requests, track accumulated effort and draft paid-addendum responses for consultants.

GPT-5.6 Sol Ultra produces proof of the Cycle Double Cover Conjecture [pdf]

GPT-5.6 Sol Ultra has produced a formal proof of the Cycle Double Cover Conjecture, marking a significant breakthrough in graph theory research.