Related papers: What Iterated Self-Feeding Probes of Language Models Measure, and a test that separates the construction from the model

What Iterated Self-Feeding Probes of Language Models Measure, and a test that separates the construction from the model

URL: http://arxiv.org/abs/2608.10986v1
Date: Tue, 11 Aug 2026 14:40:35 GMT
Title: What Iterated Self-Feeding Probes of Language Models Measure, and a test that separates the construction from the model
Abstract summary: A class of methods probes a language model by feeding it its own output.<n>We ask what such a probe measures, in a construction chosen to make the question sharp.<n>The answer is that it measures two different things at once, in readings that look alike.
Score: 0.0
License: http://creativecommons.org/licenses/by/4.0/
Abstract: A growing class of methods probes a language model by feeding it its own output: self-consistency, iterated refinement, agentic loops. We ask what such a probe measures, in a construction chosen to make the question sharp: a ring of token cells resampled in place by the model's own windowed conditional p_r(x_i | x_{i+-r}). The substrate is Glauber dynamics on token sequences and is not new; what we change is the coupling. Advancing two rings that differ in one token under common random numbers makes undamaged copies diverge by exactly zero, so damage spreading becomes measurable where a maximal coupling gives mixing times instead. The answer is that it measures two different things at once, in readings that look alike. Some quantities are fixed by the construction: the damage light cone is kinematic, and the radius scaling of the token-space Lyapunov exponent lambda_ca(r) is model-invariant across 19 models and two scale ladders spanning 70x. Others genuinely track the model: lambda_ca crosses zero at a reproducible point in training, and the attractor share ranks models consistently however the lattice is built. Left undistinguished, the first kind is readily mistaken for the second -- we did so ourselves for four months, and report a phase transition we measured to three decimal places that belongs to the probe rather than to any language model. We give the test that separates them: hold the construction fixed and vary the model, or hold the model fixed and vary the construction, and see which readings move. We validate the instrument by reproduction first, recovering a Domany-Kinzel damage field bit-exactly against an independent prediction, and we report the estimator failures that this discipline caught -- four retracted verdicts, each on a quantity that looked like a measurement. The methodology ships as a package.

Related papers

Prompt-Model Interaction Reaches the Fixed Points: A deterministic, task-free structural readout -- and the factorizations of it that failed [0.0]
We ask a readout with no task in it: the fixed-point structure of the short-window argmax map x_t+1 = argmax_x p(x | x_t-1, x_t), censused from 96 starts.<n>It is deterministic, so nothing can be helped or hurt, and it exists only at short windows -- four of six models lose it entirely by window 16 -- so everything here concerns how a model reads a fragment.<n>First, the interaction reaches this readout at full magnitude: nine tokens of conditioning move the fixed-point fraction across most of its
arXiv Detail & Related papers (2026-08-21T17:25:18Z)
Split the Labor: Separating Evidence Interpretation from Decision Aggregation [0.0]
Systems that ask a language model to reach a conclusion from many sources usually conflate them into one prompt.<n>We propose a four-field evidence (hypothesis, reliability bucket, rationale, provenance) and show that fixing it determines both halves.<n>The fix is arithmetic rather than architectural, and applies to a class of rules beyond language models.
arXiv Detail & Related papers (2026-08-14T17:24:55Z)
Intrinsic Structure: Spectral Identifiability for Mechanistic Interpretability [51.56484100374058]
We prove the first identifiability theorem for a mechanistic-interpretability primitive.<n>The spectrum is an identifiable, model-intrinsic fingerprint with a stated error bar, not a legible decomposition.
arXiv Detail & Related papers (2026-08-10T19:42:01Z)
Latent-Kernel Discrete Flow Maps for Few-Step Generation [4.159429302112398]
We build a flow-map kernel that is a mixture of M factorized components tied by a single latent.<n>We show that a single step places mass on correlated completions with the same sampling time complexity as a factorized model.<n>Experiments for unconditional text generation on the One-Billion-Word (LM1B) and WikiText-103 benchmarks show that our LKF model learns strongly heterogeneous components.
arXiv Detail & Related papers (2026-07-29T23:46:31Z)
Two Confounds in Cross-Model Value Comparison: Response Determinism and the Access Harness [0.0]
Cross-model comparisons read divergence in value dispositions as evidence that language models hold individuated values.<n>We introduce a separation protocol -- no-rule value dilemmas with counterbalanced, repeated forced-choice measurement and a determinism index.<n>We find determinism varies substantially (0.66-0.95 among engaging models)
arXiv Detail & Related papers (2026-07-11T08:26:48Z)
The Geometry of Forgetting: Temporal Knowledge Drift as an Independent Axis in LLM Representations [50.43168858368539]
Large language models confidently produce outdated answers, and no existing method can detect them.<n>We show this is not an engineering failure but a structural one: temporal drift is encoded as a direction in the residual stream geometrically to both correctness and uncertainty.
arXiv Detail & Related papers (2026-05-09T22:27:31Z)
How Focused Are LLMs? A Quantitative Study via Repetitive Deterministic Prediction Tasks [0.9338697277815541]
We investigate the performance of large language models on repetitive deterministic prediction tasks.<n>Our experiments reveal a sharp double exponential drop beyond a characteristic length scale.<n>This indicates that the models fail to execute each operation independently.
arXiv Detail & Related papers (2025-11-02T01:42:08Z)
Independence Tests for Language Models [47.0749292650885]
Given the weights of two models, can we test whether they were trained independently?<n>We consider two settings: constrained and unconstrained.<n>We propose a new test which matches hidden activations between two models, and which is robust to adversarial transformations and to changes in model architecture.
arXiv Detail & Related papers (2025-02-17T20:01:08Z)
Gumbel Counterfactual Generation From Language Models [64.55296662926919]
We show that counterfactual reasoning is conceptually distinct from interventions.<n>We propose a framework for generating true string counterfactuals.<n>We show that the approach produces meaningful counterfactuals while at the same time showing that commonly used intervention techniques have considerable undesired side effects.
arXiv Detail & Related papers (2024-11-11T17:57:30Z)
Is Model Collapse Inevitable? Breaking the Curse of Recursion by Accumulating Real and Synthetic Data [49.73114504515852]
We show that replacing the original real data by each generation's synthetic data does indeed tend towards model collapse. We demonstrate that accumulating the successive generations of synthetic data alongside the original real data avoids model collapse.
arXiv Detail & Related papers (2024-04-01T18:31:24Z)
How Can We Know When Language Models Know? On the Calibration of Language Models for Question Answering [80.82194311274694]
We examine the question "how can we know when language models know, with confidence, the answer to a particular query?" We examine three strong generative models -- T5, BART, and GPT-2 -- and study whether their probabilities on QA tasks are well calibrated. We then examine methods to calibrate such models to make their confidence scores correlate better with the likelihood of correctness.
arXiv Detail & Related papers (2020-12-02T03:53:13Z)

This list is automatically generated from the titles and abstracts of the papers in this site.