TLDR Tech / Practical
I Used OpenAI Dots as an Agent Swarm to Break a 47 Year Old Math Record For Free (with Lean Verification of the Proof)
The author used an OpenAI Dots agent swarm of seven subagents to solve a math covering problem, proving that C(24,14,4) ≥ 20 and raising the bound for C(25,15,5) to 34. The swarm divided work into mathematical research, brute forcing, adversarial review, and Lean formalization, which passed mechanical checks at the Palomar registry. The author learned that verification is critical because agents are confident yet prone to errors, and that task pool architectures help manage agents when compute is free and unlimited.
Read the full story