多智能体大语言模型系统中的林格曼效应:有效团队规模的缩放定律
The Ringelmann Effect in Multi-Agent LLM Systems: A Scaling Law for Effective Team Size
AI总结:
本文推导出两参数缩放定律 $R(N) = N_\text{eff}/N = 1/(1+c(N-1)N^{-\beta})$,将多智能体LLM系统分为三种渐近状态,并通过44个实验单元验证了该定律,发现密集辩论无法增加答案多样性,噪声安慰剂可模拟自我修正效果,且仅异构团队能突破硬上限。
摘要:
推理时多智能体大语言模型缩放缺乏共享单位:计数名义智能体混淆了成本与独立证据。我们推导出一个两参数缩放定律 $R(N) = N_\text{eff}/N = 1/(1+c(N-1)N^{-\beta})$,其中状态指数 $\beta$ 将任何配置分类为三种渐近状态之一——硬上限为 $1/c$($\beta = 0$)、亚线性为 $N^\beta/c$($0 < \beta < 1$)或线性($\beta \ge 1$),并且平均场定理预测智能体辩论中的同伴数量 $k$ 和轮次 $\tau$ 仅通过其乘积 $k\tau$ 进入动力学。该定律适用于两个层面:答案多样性和正确性冗余。在44个(模型 $\times$ 任务 $\times$ 条件)单元中,涵盖同伴辩论、自我修正、随机噪声安慰剂、自一致性、三个开放权重系列(Qwen、Llama、Ministral)从7B到32B规模,并辅以前沿API检查(Gemini)、思维模型、异构团队和稀疏通信,函数形式在每个条件下拟合 $R^2 > 0.99$;仅 $(c, \beta)$ 发生偏移。在自由形式数学问题上,密集同伴影响将答案层面状态从亚线性坍缩为硬上限;正确性层面拟合始终保持硬上限。三个发现具有实际意义。 (i) 三十个密集辩论智能体在MMLU-Hard上产生的答案多样性不超过一个智能体。 (ii) 噪声安慰剂在自由形式数学问题及4倍规模下追踪自我修正,因此在同质团队中,通常归因于“辩论”的收益来自重新评估,而非同伴内容。 (iii) 单个 $N \le 5$ 的试点预测了 $N=30$ 的结构上限,并且在测试的配置中,只有架构多样性(异构团队)降低了 $c$ 并逃离了硬上限状态,通信模式干预则不能。
英文摘要:
Inference-time multi-agent LLM scaling lacks a shared unit: counting nominal agents conflates cost with independent evidence. We derive a two-parameter scaling law $R(N) = N_\text{eff}/N = 1/(1+c(N-1)N^{-β})$ where the regime exponent $β$ classifies any configuration into one of three asymptotic regimes -- hard-ceiling at $1/c$ ($β= 0$), sublinear at $N^β/c$ ($0 < β< 1$), or linear ($β\ge 1$), and a mean-field theorem predicts that peer count $k$ and rounds $τ$ during agent debate enter the dynamics only through their product $kτ$. The law applies at two levels: answer diversity and correctness redundancy. Across 44 (model $\times$ task $\times$ condition) cells spanning peer debate, self-correction, random-noise placebo, self-consistency, three open-weight families (Qwen, Llama, Ministral) at scales from 7B to 32B with a frontier API check (Gemini), thinking models, heterogeneous teams, and sparse communication, the functional form fits every condition at $R^2 > 0.99$; only $(c, β)$ shifts. On free-form math, dense peer influence collapses the answer-level regime from sublinear into hard-ceiling; correctness-level fits remain hard-ceiling throughout. Three findings have practical implications. \emph{(i)}~Thirty dense debating agents produce no more answer diversity than one on MMLU-Hard. \emph{(ii)}~A noise placebo tracks self-correction on free-form math and at $4\times$ scale, so within homogeneous teams the gain commonly attributed to ``debate'' comes from re-evaluation, not peer content. \emph{(iii)}~A single $N \le 5$ pilot predicts the $N=30$ structural ceiling, and within the configurations tested only architectural diversity (heterogeneous teams) lowers $c$ and escapes the hard-ceiling regime, communication-mode interventions do not.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。