Harvard's MPT trains decentralized LLM agent swarms to self-orchestrate via learned bids
du_yilun · x · 2026-10-20
Harvard researchers (Guowei Xu, Zhenting Qi, Yilun Du et al.) introduce Market-Based Post-Training (MPT), inspired by Society of Mind, exploring whether intelligence can emerge from decentralized interactions among specialized LLM agents rather than scaling a single model.
- MPT jointly learns action policies and self-orchestration: agents bid to express willingness to participate; bids determine who acts, and collective outcomes provide local training credit — no central coordinator
- Evaluated on embodied crafting, robot navigation, and deep research, outperforming monolithic post-training; learned self-orchestration beats alternative coordination strategies
- Ablations show both acting and bidding benefit from learning; trained swarms adapt faster to related tasks, scale better at inference, and degrade less under agent dropout
More from coding & agent
- OvertChat: a polished self-hosted ChatGPT-style UI for local models, one-command install — anitamaxwynnn69 · 2026-10-21
- Smart Connections demo: Obsidian surfaces past decisions to reshape your plan — AINewsletter · 2026-10-21
- Ampersand v0.1 Ships via Agent API: Hosted Sessions, OpenAI-Client Compatible — gharik · 2026-10-21
- Pi Durable: Mario Zechner and Armin Ronacher on crash-safe long-running agents — mitsuhiko · 2026-10-21
- POV: Everyone's Too Busy vibe-coding Their Own Game to Play Yours — yacineMTB · 2026-10-21
- Grokbot agent drafts industrial control schematics in open-source PanelCAD — burhop · 2026-10-21