Noam says multi-agent deserved under 10% credit for Navier-Stokes solve, sparking debate
scaling01 · x · 2026-09-19
Elie Bakouch and scaling01 clashed over how much credit the multi-agent setup deserves for the AI-assisted Navier-Stokes Millennium Prize proof.
- Noam's quote: solving the problem "was not due to multi-agent" — he wouldn't even attribute 10% of the credit to the agent swarm.
- Bakouch reads this as multi-agent contributing under 10%, noting Noam called it one data point with no single-agent control experiment run yet (a full ablation would be too expensive).
- scaling01 counters: math is highly parallelizable with slightly sublinear scaling, implying a single model would have needed years — the proof wouldn't exist now without multi-agent scaling — so the generational model uplift matters less than the swarm, making the 10% framing untenable.
More from coding & agent
- Microsoft ships Playwright Workspaces remote MCP with 22 browser tools for agents — WirelessLife · 2026-09-19
- Ada.Url brings the Ada URL parser to .NET: 1.9x faster than System.Uri, WHATWG-compliant — lemire · 2026-09-19
- DiffUI Creator: Explore Designs With Images First, Then Hand Them to Coding Agents — msg · 2026-09-19
- 76% of orgs ship AI coding assistants but only 34% can measure impact — activity metrics are misleading — rseroter · 2026-09-19
- Xcode can export skills for coding agents, including a full app audit command — rudrank · 2026-09-19
- Xcode 27.1's 'Get my app ready for iPhone Duo' Runs a 7-Point App Audit — rudrank · 2026-09-19