内部机制不可解,只刻画输入到输出的映射
The mechanism is intractable; characterize only the input–output map
A linear-response lens on LLM agents
预算越大、问题越复杂,固定的 LLM 越落后于固定策略:每多一份预算,它换来的收益更少。
The larger the budget and the more complex the problem, the further a fixed LLM falls behind a fixed strategy: each extra unit of budget buys it less.
01 / MOTIVATION
LLM 有数千亿参数,内部动力学无从逐一追踪。统计物理面对复杂系统,不求解微观细节,只施加小扰动、测量宏观响应。我们对 LLM 采取同样的立场。
An LLM has hundreds of billions of parameters; its internal dynamics cannot be traced one by one. Statistical physics does not solve the microscopic details of a complex system. It applies a small perturbation and measures how a macroscopic quantity responds. We take the same stance toward the LLM.
内部机制不可解,只刻画输入到输出的映射
The mechanism is intractable; characterize only the input–output map
预算的微小增量引起效用的增量; 可测,且与微观实现无关
A small budget increment produces a utility increment; is measurable and independent of the microscopic realization
固定 LLM 串入流水线,把固定策略的响应率 变为
A fixed LLM inserted in the pipeline turns the base response into
只测响应、不问机制:经过固定 LLM 后,响应率会变大还是变小?
Measure only the response, not the mechanism: does a fixed LLM raise or lower the susceptibility?
02 / DERIVATION
固定策略在预算 下给出策略集 ;固定 LLM 读入后改写为 。改写的效果有两面。
A fixed strategy gives a strategy set at budget ; a fixed LLM reads it and rewrites it as . The rewrite cuts both ways.
基线离最优尚远时,LLM 的事实与常识足以纠错,衍生策略领先。
While the baseline is far from optimal, the facts and common sense the LLM holds can correct it, and the derived strategy leads.
基线接近最优后,改进空间收窄,LLM 的改动更可能引入错误。
Once the baseline is near optimal, the room for improvement shrinks and the LLM's changes are more likely to add errors.
LLM 参数固定,其信息贡献不随预算增长;固定策略却随预算不断逼近最优。
With fixed parameters, the LLM's information does not grow with budget, while the fixed strategy keeps closing in on the optimum.
预算足够大时,经过固定 LLM 的响应率 不超过固定策略的 。
At large budget, the susceptibility through a fixed LLM does not exceed the fixed strategy's .
上界只约束参数固定的 LLM;让 LLM 随预算一同扩大,才能越过它。
The bound constrains only a fixed-parameter LLM. Letting the LLM itself grow with the budget is the way past it.
参数固定的 LLM 可视为容量有限的信道。由数据处理不等式,衍生集与最优策略的互信息不超过基础集与最优策略的互信息。该论证给出动机,并非证明;“足够大”的位置由实验确定。
A fixed LLM acts as a channel of finite capacity. By the data-processing inequality, the derived set carries no more mutual information with the optimum than the base set. This motivates the bound but does not prove it; where “sufficiently large” begins is set by experiment.
03 / EXPERIMENTS
上界给出两条可检验的预言:LLM 的领先随预算消失、落后随预算扩大;任何固定配置的 都不超过 1。
The bound makes two testable predictions: the LLM's lead converges and its deficit widens with budget; no fixed configuration has significantly above 1.
三个任务连成一条弧线:LLM 从领先,到持平,再到落后。金色阴影为 LLM 领先,蓝色为固定策略领先。
Three tasks form one arc: the LLM goes from ahead, to level, to behind. Gold shading marks an LLM lead, blue a fixed-strategy lead.
从带噪估计中找出第一名。信噪比低时,LLM 凭常识领先;信噪比升高后,算法收敛到真值,领先消失。
Pick the top item from noisy estimates. At low SNR the LLM leads on common knowledge; as SNR rises the algorithm converges to the truth and the lead vanishes.
生成器给出 个解答,由多数投票或固定 LLM 选择器选定答案。样本少时两者持平;样本增多后,选择器明显落后。
A generator writes attempts; majority vote or a fixed LLM selector picks the answer. With few samples the two are level; with more, the selector falls clearly behind.
LLM 在束搜索的前三个候选中择一。启发式基线本已很强,LLM 从一开始就落后,差距随束宽扩大。
The LLM picks among beam search's top three candidates. The heuristic baseline is already strong, so the LLM trails from the start, and the gap widens with beam width.
LLM 曲线为多个模型的平均。
LLM curves average several models.
对每个固定配置,取 LLM 与固定策略的斜率之比 。
For each fixed configuration, take the ratio of the LLM slope to the fixed-strategy slope.
04 / APPLICATIONS
在经过固定 LLM 的通道上追加预算,收益追不上固定策略。值得投入的有两处:让 LLM 随预算一同扩大,或投向边际响应仍为正的通道。
Extra budget on a channel through a fixed LLM cannot catch the fixed strategy. Two places are worth the budget: let the LLM grow with it, or spend on channels whose marginal response is still positive.
一种常见的回答问题方法是先生成多个选项,然后从中选择最好的一个。保持选择器不变,逐步提升生成器的能力,就能考察更强的生成器对最终答案正确率的影响;此时固定选择器仍受响应率上界约束。如果让选择器与生成器使用同一个模型、一起变大,选择器本身就随预算增长,上界不再适用。
A common way to answer a question is to generate several candidate answers, then select the best one. Keeping the selector fixed while increasing the capability of the generator shows how a stronger generator affects the accuracy of the final answer. The fixed selector still obeys the susceptibility bound. If the selector and generator use the same model and grow together, the selector itself grows with the budget, so the bound no longer applies.
示意
Schematic
嵌套从最低处出发,到 Qwen3-Max 时追平多数投票,斜率在所有配置中最陡。
Nesting starts lowest and reaches majority vote at Qwen3-Max: the steepest slope of all configurations.
在一个真实的 agent 任务中,agent 面对二维的选择:用多少工具,交互多少轮。在这个二维空间中,响应率决定了任务得分随着选择变化的整体规律。
In a real agent task, the agent must choose along two dimensions: how many tools to use and how many turns to take. In this two-dimensional space, susceptibility determines the overall pattern of how the task score changes with these choices.
响应曲面 由所选模型的 15 条轨迹检查点平滑拟合。白线标出各轮次拟合评分最高的工具数;金色虚线跟随轮次控件。
The response surface is a smooth fit to checkpoints from the selected model's 15 trajectories. The white line marks the tool count with the highest fitted score at each turn; the gold dashed line follows the turn control.
要紧的不是最优点落在何处,而是曲面的形状:每条预算通道的边际响应终将趋零或转负,且仅凭黑箱观测即可测得。
What matters is not where the sweet spot sits but the shape of the surface: every channel's marginal response eventually flattens or turns negative, and this is measurable from black-box observations alone.
示意
Schematic
为此,可以设计一个外置优化器。优化器在运行中做同一件事:以各轨迹的检查点在线估计响应曲面及其不确定度,把后续预算投向边际响应最高处,响应已趋零的方向不再投入。它只读取黑箱观测。
For this purpose, we can design an external optimizer. The optimizer does the same thing while it runs: from the checkpoints of all trajectories it estimates the response surface and its uncertainty online, sends further budget where the marginal response is highest, and none where it has flattened. It reads only black-box observables.
Agent 的目标是在有限 token 预算内,根据物理论文生成尽可能多的合格推导题,每题附答案和完整解答。合格题须有论文依据、需要多步推理,并通过独立求解与原文核查。这里以五篇《Physical Review Letters》(PRL)论文为素材,涵盖量子空间搜索、量子行走复苏、渗流图上的行走与光子广义测量。
The agent aims to generate as many good physics derivation questions as possible within a limited token budget, each with an answer and a complete solution. A good question must be supported by the paper, require multi-step reasoning, and pass an independent solve and a check against the source paper. The inputs are five Physical Review Letters (PRL) papers on quantum spatial search, quantum-walk revival, walks on percolation graphs, and photonic generalized measurements.
05 / DISCUSSION
三个延伸:以零响应率为基线时的上界,多个异质通道同时经过 LLM 时的例外,以及对自我演化的含义。
Three extensions: the form of the bound against a zero-susceptibility baseline, its limits when several heterogeneous channels pass through the LLM, and what it implies for self-evolution.
每一步续写都是固定 LLM 作用于已有生成;对照是原样返回输入、响应率为零的处理器。上界因此预言:链足够长后,正确率随链长下降。
Each continuation step is a fixed LLM acting on the generation so far; the reference is a processor that returns its input, with zero susceptibility. The bound therefore predicts that accuracy falls with chain length once chains are long enough.
四个模型在 PHYBench 上使用不同题目的 TTS 估计。链长除以同一模型在该题上的中位链长,正确率采用对数刻度。点为分箱估计,线为指数拟合,阴影为 1σ 拟合带;最短区间的淡色点不参与拟合。
TTS estimates for four models using different PHYBench questions. Chain length is divided by the median length for the same model and question. Accuracy uses a logarithmic scale. Points show bin estimates, lines show exponential fits, and shading shows the 1σ fit band. The pale point in the shortest interval is excluded from the fit.
观察性关联。
An observational association.
以上实验每次只改变一个通道。多个通道同时经过 LLM 时,每个通道各自有损、各受上界约束。
The experiments above vary one channel at a time. When several channels pass through the LLM together, each is lossy and bounded on its own.
示意
Schematic
信号异质时,比较随之改变:多模态观测须合并为一次推断时,常规代码往往无从结合,只能利用其中一路。
The comparison changes when signals are heterogeneous. When multimodal observations must be combined into one physical or engineering inference, conventional code may have no way to combine them, so its response to the joint signal comes only from the channel it can process.
此时 LLM 虽逐通道有损,仍能利用代码无法利用的信息;在极端或开放环境中,agent 可能胜过代码。
A lossy LLM can then still use information the code cannot; in extreme or open-ended environments, the agent may outperform the code.
假设能力为 的模型生成训练数据,其质量为 ;用这批数据训练后,模型的输出质量 升至数据的水平,能力随之更新,如此往复。 与 谁增长更快,决定这条阶梯的去向。
Assume that a model of capability generates training data of quality ; after training on it, the model's output quality rises to match the data, and its capability moves to a new b. Then it repeats. Whether or rises faster decides where the staircase goes.
固定的 LLM 追不上固定策略的斜率;
唯有随预算一同成长,才能越过这条线。
A fixed LLM cannot escape the slope of a fixed strategy; only by growing with the budget can it cross the line.