Key points
- OpenAI published solutions or partial solutions to 372 outstanding mathematical challenges.
- The work was generated by an internal, unreleased AI model and spans 722 papers.
- Mathematician Terence Tao and the Association for Human Mathematics called for colleagues to abandon projects with OpenAI.
OpenAI has released solutions to 372 outstanding mathematical problems, prompting significant criticism from the academic community. Experts argue that the scale and complexity of the output undermine traditional scholarly verification and human understanding of the field.
What happened
OpenAI published solutions or partial solutions to 372 mathematical challenges that have remained unsolved by human researchers for decades. According to TechRadar, these outputs were generated using an internal, unreleased AI model designed specifically for advanced mathematics. The release consists of 722 papers, a volume that TechRadar notes makes it difficult for human mathematicians to verify the results.
The solutions are described as nearly as complex as the problems themselves. This complexity creates a barrier to verification, as the methods used by the AI are not always transparent or understandable to human experts. The publication follows a recent proposal by OpenAI regarding the Navier-Stokes equation, which concerns the behaviour of viscous fluids.
Why it matters
Terence Tao, a renowned mathematician, has called for colleagues working with OpenAI to abandon their projects. He shared a statement from the Association for Human Mathematics, which argues that releasing over 700 files simultaneously demonstrates power rather than scholarship. The association urges a return to a vision of science that centres on human understanding.
The Association for Human Mathematics states that mathematicians did not request this work to be done. They highlight that the Advisory Group on Mathematics and Artificial Intelligence, from whom OpenAI claims legitimacy, previously advised against testing advanced problems on internal models. This contradiction suggests a misalignment between industry actions and academic guidance.
The rapid advancement is stark when compared to just two years ago, when AI struggled with GCSE-level mathematical challenges. The shift from basic arithmetic to solving decades-old advanced problems indicates a significant capability jump. However, this speed has come at the cost of interpretability, raising concerns about the long-term impact on mathematical research.
What happens next
- Monitor official statements from the Association for Human Mathematics regarding further guidance.
- Review internal policies on collaboration with AI firms for research integrity.
- Assess the verification capabilities of current teams for AI-generated mathematical proofs.
- Await potential regulatory or academic responses to the mass publication of solutions.
What to do and how to stay safe: OpenAI
- Establish clear criteria for verifying AI-generated outputs in technical research.
- Ensure that any AI-assisted work remains interpretable by human experts.
- Review collaboration agreements to align with academic ethical guidelines.
- Monitor updates from relevant advisory groups on AI usage in mathematics.
Step-by-step guide: Shadow AI usage: how to control unmanaged model access in your firm
General security guidance from the AppSec Scoop newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
How many mathematical problems did OpenAI solve?
OpenAI released solutions or partial solutions to 372 outstanding mathematical problems.
Who is leading the opposition to OpenAI's release?
Renowned mathematician Terence Tao is leading the opposition, supported by the Association for Human Mathematics.
What model was used to generate these solutions?
The solutions were produced using an internal, unreleased AI model specialised in mathematics.
