Using OpenAI Dots as an agent swarm to prove new math, for free (Lean verified) (unexcitedneurons.substack.com)

OpenAI recently launched Dots, a personal assistant powered by GPT-6 Astra agents, which offers free access to a swarm of AI agents for various tasks, including complex mathematical problem-solving. Utilizing an AMD Epyc CPU with 9 cores and 10GB of RAM, the Dots agent successfully proved that the covering number C(24,14,4) is at least 20, improving upon previous lower bounds established in the 1960s. The computation took three days of processing power and utilized a collaborative approach where seven agents managed different roles—such as mathematical research, formal verification, and computational search— to tackle the proof effectively. This development is significant for the AI/ML community as it showcases the potential of collaborative agent architectures for tackling intricate problems in mathematics, highlighting not only efficiency but also innovative approaches to verification within AI systems. The ability to engage in such high-level mathematics at zero cost demonstrates the democratization of advanced computational tools, paving the way for broader engagement in mathematical research. The Dots agents’ orchestration involved switching from fixed roles to a shared task pool, allowing dynamic task allocation and promoting more effective collaboration, lessons that could be pivotal for future AI-assisted research endeavors.