Researchers Spar Over Whether 10,000-Agent Parallel Scaling Beats New Models

A technical debate has broken out among researchers over how much credit an AI agent swarm deserves for the discovery of a Navier-Stokes proof. One claim put the swarm's contribution at roughly 10%, but posters including @scaling01 pushed back, calling that claim flat-out wrong. TL;DR: the argument centers on the relative contributions of parallel test-time compute (parallel TTC) versus serial reasoning, with critics arguing that without scaling parallel TTC to 10,000 agents, the proof simply would not exist.

Confirmed

Unconfirmed

Why it matters

2026-09-18 ~ 2026-09-19 · 9 related posts

Primary sources

1 near-duplicate retellings: scaling01