Breakthrough in Computational Complexity via AI Research
Anthropic has reportedly utilized an internal research model to resolve long-standing mathematical hypotheses, marking a significant development in the application of large language models (LLMs) to theoretical computer science. According to reports circulating on social media platforms like Reddit, the model successfully identified a new algorithm that provides polynomial improvements over established complexity bounds for two fundamental problems: the 3SUM (3-Sum) and APSP (All-Pairs Shortest Path) hypotheses.
The findings, which have been shared with academic researchers, suggest a deterministic complexity of $O(n^{1.9992})$ for 3SUM and $O(n^{2.9995})$ for integer-weight APSP. These figures represent the first polynomial improvements over the traditional $n^2$ and $n^3$ textbook bounds. The core of the discovery involves a new thin matrix product algorithm, which researchers suggest could lead to speedups in related problems such as Exact Triangle, Zero-Weight k-Clique, and Tree Edit Distance.
Methodology and Verification
The discovery process was characterized by a high degree of autonomy. Anthropic reported that an employee tasked the internal model with verifying and improving cryptographic constructions based on the average-case hardness of Zero-k-Clique. Instead of merely verifying the existing work, the model developed the new algorithm for both average and worst-case scenarios.
The session reportedly utilized 16 million output tokens without human input. To ensure mathematical rigor, the main theorems were formalized in Lean, a formal proof language. Anthropic has shared the algorithm with the relevant authors under a confidentiality agreement as of September 2026, providing compensation and access to the public version of its Claude model.
Academic and Ethical Debates
The development has ignited a debate within the mathematical community regarding the role of AI in high-level research. Some specialists express concern over “cognitive technical debt,” arguing that AI-generated proofs can be massive, bloated, and difficult for humans to parse. Critics suggest that while an AI might provide a “Yes/No” answer to a theorem, it may fail to provide the deep conceptual understanding that defines traditional mathematics. There are concerns that if AI continues to produce “sloppy” or overly complex proofs, human mathematicians will be left to “pick up the pieces” to extract meaningful theory from the output.
Conversely, some proponents argue that the ability of AI to solve frontier questions is a matter of efficiency rather than a loss of value. They suggest that if AI can lead to practical scientific advances, the method of discovery—human versus machine—is secondary to the result.
Economic and Structural Implications
The ability of AI to tackle cognitive tasks rather than just manual or repetitive ones has raised questions regarding the future of highly skilled labor. While some observers view these advancements as a way to democratize science by providing tools to those without massive computing resources, others warn of a growing divide. There is concern that scientific research could become monopolized by a few well-funded institutions capable of maintaining the necessary compute power.
As AI moves from automating implementation to automating understanding, the impact on professional sectors remains a subject of intense speculation. While the immediate mathematical implications are theoretical, the long-term economic consequences of automating cognitive labor remain unquantified.