Claims of AI Math Breakthroughs Lack Verifiable Evidence, Sources Show — E8 Intelligence Research
Year: 2026
Type: preprint
Abstract: FINDING: DARPA's expMath program and OpenAI's "Astra" model claim 10 solved mathematical challenges, but the provided sources are video titles/abstracts with no actual mathematical content disclosed. The only concrete technical item is a multi-robot exploration system for the DARPA Subterranean Challenge (arXiv:2110.05911), which is engineering, not pure mathematics. MATH: No equations, constants, or ratios are extractable from the given evidence. The arXiv paper describes SLAM, path planning, and multi-agent coordination — no novel mathematical formalism is presented in the abstract. The OpenAI claims (Collatz formalization, etc.) are mentioned but not substantiated with any specific result, proof, or constant. CONNECTION: None directly supported. However, the Subterranean Challenge inherently involves lattice/graph structures (topological maps of tunnels), and multi-robot coordination often exploits symmetry groups (e.g., permutation symmetries in task allocation) and root-system-l Author: Andrew Stewart Caldin, Independent Researcher, UK. Part of the E8 Intelligence Research series. Platform: e8intelligence.com
Author: Caldin, Andrew Stewart
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7212435010
Cites:
Cited by:
Related to: