发表机构
Chengdu Beiluoshimen Technology Co., Ltd.（成都北落师门科技有限公司）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
Lingtai是一种无需训练的概念遥测层，通过投影残差状态到概念锚点来观察LLM推理，揭示与不确定性相关的活动信号和轨迹身份，但正确性需外部提供，开销极小。
AI 中文摘要
在自回归推理过程中观察大型语言模型计算的内容——在线且无需训练探针——仍然困难。我们引入了Lingtai，一个无需训练的概念遥测层：在每个生成步骤，残差状态被投影到一组特定领域的命名概念锚点上，这些锚点是在没有标记概念示例、结果标签、梯度拟合或激活空间优化的情况下构建的，从而产生结构化的逐步骤概念坐标信号。在代码生成和小学数学推理中，该信号与预测不确定性表现出稳健的关联：这种关联在控制问题身份和令牌位置后依然存在，不能归因于单一令牌类型，不能由GSM8K上的简单正确/错误混合解释，也不能由匹配的随机锚点复现；在K-means和PCA投影中，这种关联明显较弱或方向不一致。出现了两种结构：一种反复出现的与不确定性相关的活动信号，其功能几何是任务条件化的（在HumanEval、MBPP和GSM8K上具有不同的活动熵形状），以及一种执行特定的轨迹身份，具有强局部惯性但弱的重新实例化不变性——在仅完成弹性对齐下，对匹配的重新执行子集在k=32（大约完成的中位数四分之一）处的损坏仍以62.0%的概率检索到存档片段，而新执行仅以11.7-16.0%的概率检索到。最后，匹配的审计发现没有证据表明此处使用的标量概念活动信号在测试协议下提供稳定的正确性坐标；因此我们将正确性视为外部提供的。遥测为161锚点的代码实现增加了每令牌0.7-1.6%的解码开销，且生成的令牌不变。
英文摘要
Observing what a large language model computes during autoregressive inference--online and without training probes--remains difficult. We introduce Lingtai, a training-free concept telemetry layer: at each generation step, residual states are projected onto a domain-specific bank of named concept anchors, constructed without labeled concept examples, outcome labels, gradient fitting, or activation-space optimization, producing a structured per-step concept-coordinate signal. Across code generation and grade-school mathematical reasoning, this signal exhibits a robust association with predictive uncertainty: the association survives problem-identity and token-position controls and is not attributable to a single token type, is not explained by a simple correct/incorrect mixture on GSM8K, and is not reproduced by matched random anchors; it is markedly weaker or direction-inconsistent in K-means and PCA projections. Two structures emerge: a recurring uncertainty-linked activity signal whose functional geometry is task-conditioned (distinct activity-entropy shapes on HumanEval, MBPP, and GSM8K), and an execution-specific trajectory identity with strong local inertia but weak re-instantiation invariance--under completion-only elastic alignment, corruption at k=32 (approximately a median quarter of the completion) on the matched re-execution subset still retrieves the archived episode at 62.0%, while a fresh execution retrieves it only 11.7-16.0% of the time. Finally, a matched audit finds no evidence that the scalar concept-activity signal used here supplies a stable correctness coordinate under the tested protocol; we therefore treat correctness as externally supplied. Telemetry adds 0.7-1.6% per-token decode overhead for the 161-anchor code implementation, with unchanged generated tokens.
Comments15 pages, 4 figures, 7 tables. An earlier version was publicly released on Zenodo (DOI: 10.5281/zenodo.23068698)
