当代理过早承诺时：如何识别大语言模型代理的过早承诺问题
When Agents Commit Too Soon: Diagnosing Premature Commitment in LLM Agents
摘要
长视距LLM智能体可能会悄无声息地失败：它们最初就选定了某种对证据的解读方式，然后在整个运行过程中都坚持这种解读。我们将其称为“过早承诺”。最终答案评分无法识别这种失败模式，因为它只关注答案本身，而不考虑该过程是否已经稳定到某个状态。我们将“表征性承诺”定义为在固定推理步骤中隐藏状态的收敛现象，并将其作为判断路径一致性的早期指标。在Llama-3.1-70B上使用ReAct技术处理HotpotQA任务时，第4步的隐藏状态相似性可以预测后续行为的一致性（r = -0.35，部分r = -0.45），同时还能识别出特定的时间性和层级特征。这一信号在Qwen-2.5-72B和Phi-3-14B上同样有效，在StrategyQA任务中则表现为r = -0.83。不过，它无法反映正确性：那些被错误或正确识别的问题在激活相似性上无法区分开来。这个边界点对于判断智能体的状态至关重要。所谓“承诺”，指的是智能体是否已经确定了某种立场，而不是其判断是否正确。运行时监控工具能够检测到隐藏状态之间的不一致性，其AUROC值可达0.97（在更严格的分类下为0.85–0.88）。而通过提示词干预措施，可以將行为差异降低28%，而准确性则保持不变。我们还测试了这一信号是否能够帮助实现自我一致性计算；在更复杂的基准测试中，它的作用较为有限，可以被更简单的基于输出的基线方法所替代。因此，这一指标可以用来诊断隐藏过程中的故障，但它有明确的限制，而非一种通用的准确性提升手段。
English Abstract
Long-horizon LLM agents can fail quietly: they settle on one reading of the evidence early, then spend the rest of the run defending it. We call this premature commitment. Final-answer scoring misses the failure mode because it sees only the answer, not whether the process has already collapsed to a stable path. We define representational commitment as cross-run hidden-state convergence at a fixed reasoning step, and use it as an early diagnostic of trajectory consistency. On Llama-3.1-70B running ReAct on HotpotQA, step-4 hidden-state similarity predicts downstream behavioral consistency (r = -0.35, partial r = -0.45), with a localized temporal and layer-wise signature. The signal replicates across Qwen-2.5-72B and Phi-3-14B, and on StrategyQA (r = -0.83). It does not track correctness: committed-wrong and committed-correct questions are not separable in activation similarity. That boundary is central to the claim. Commitment tells us whether an agent has settled, not whether it is right. A runtime monitor detects inconsistent trajectories from hidden states at AUROC up to 0.97 (0.85--0.88 under a stricter split), and a prompting intervention cuts behavioral variance by 28% against a token-matched control while leaving accuracy statistically unchanged. We also test whether the signal can route self-consistency compute; on a harder benchmark it helps only modestly and is matched by a simpler output-based baseline. The result is a diagnostic for a hidden process failure, with clear limits rather than a general accuracy lever.