CORTEXA
← Browse
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-23Cited by 0

Reading the substrate, not the score: a self-calibrating friction signature that warns of fine-tuning over-fitting and maps when interventions help

Tomas Pødenphant Lund

Early warning of a training-time collapse has so far been read from a model’s internals. This paper shows the same precursor is legible in the model’s own output distribution: per-token entropy variance and lag-1 autocorrelation, computed from the log-probabilities any inference interface already returns, rise before held-out accuracy drops when narrow fine-tuning begins to erode general reasoning. The two precursors replicate across three model families (Qwen2.5-1.5B, Phi-3.5-mini, SmolLM2-1.7B) with multi-seed error bars, and the lead over the accuracy drop runs from several optimizer steps to coincident depending on the model. The signature is packaged as a self-calibrating monitor whose trigger is relative to each run’s own opening baseline, so there is no global threshold and no separate calibration pass. It flags the collapse in three of three families that collapse, while a difficulty-masking gate correctly skips a fourth. The load-bearing contrast is that the failure is capability-erosion rather than over-fit to the training task: the in-task validation loss falls straight through the collapse and never warns, and a ceiling-masked accuracy probe never moves, while the friction signature rises through the transition. An estimator ablation shows the advantage of the distributional moments over a scalar entropy level is transfer rather than lead — the level matches them on two families and fails on the third. Read across model capacity instead of training time, the same signature yields a regime-map that predicts when an instruction frame helps: it bites on pressed substrates and does nothing on frozen ones. A published account of entropy collapse proves it a first-order transition for feedback-amplified systems, across which variance and autocorrelation stay finite, so that no such precursor can exist there. That result and this one concern different system classes: its defining feature, the model’s own output re-entering its input, is absent from supervised fine-tuning on a fixed external corpus. The paper states as an explicitly untested conjecture that the signature should fail on recursive self-training, and names the experiment that would decide it. The monitor’s statistical construction is inherited rather than proposed: it is a control chart with a baseline-relative limit, the Phase-I / Phase-II practice of statistical process control and change detection. What is specific here is the observable — a friction quantity read from the model’s own output distribution rather than a task metric or an internal state — and the honest limits that follow: the threshold is set to a ceiling on benign drift rather than to a false-alarm rate, and a panel of never-collapsing runs is named as the validation that would supply the guarantee sequential methods provide. Target venue: Transactions on Machine Learning Research. Data, notebooks, and analysis scripts are released with the paper. Part of the Friction Theory research programme; companion papers are linked in the related identifiers. New in version 2. The empirical results are unchanged; the revision is one of positioning. Version 1 shipped without a related-work section. This version adds one covering six literatures, narrows the novelty claim from the early-warning phenomenon (which prior work established) to the observable (that the precursor is legible in the output distribution, hence deployable without weight access), and reconciles the regime-map against its incumbents. It also positions the monitor against the statistical field that owns its construction — sequential change detection and control charts — conceding that the construction is inherited and locating the specific contribution in the observable. An estimator ablation on the existing trajectories, added here, reports the distributional moments as buying transfer across families rather than a longer lead. Reviewed by external dual peer-style review of the revision. Builds on v1.

View free PDFSource page

Related papers

openalexZenodo (CERN European Organization for Nuclear Research)2026-07-24

Controllable Generative AI: A Technical Review of Model Parameters for Hallucination Mitigation, Fine-Tuning, and Replicable Model Construction

Prateek Dutta

Generative artificial intelligence (GenAI) systems, particularly large language models (LLMs), expose a dense and interacting set of architectural, optimization, and inference-time parameters that jointly determine factual reliability, adaptation quality, and reproducibility. Thi…

View free PDFSource page
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-26

Electrical resistivity tomography surveys, trained physics-informed neural network models and code for amortized ERT inversion along Route Regionale 707, Moroccan Middle Atlas

Rajae Ajana

This deposit contains the field data, synthetic training datasets, trained network weights and analysis code supporting the article "Physics-informed neural network inversion of electrical resistivity tomography data: amortized optimization with field validation in the Moroccan M…

View free PDFSource page
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-26

Automated Detection of Self-Harm Wounds Using Deep Learning and Image Processing in Forensic Medicine

A Mohammadi, Mahdi Mehrabi, Seyed Mohammad Saadatneshan, Kamroz Amini, Mahdi Gheysari

Background and Objective: Self-harm is a psychologically damaging behavior, and its accurate differentiation from other wounds (violence, accidents, burns, diabetic ulcers) is critically important in forensic medicine. However, this differentiation often falls into a diagnostic "…

View free PDFSource page
openalexZenodo (CERN European Organization for Nuclear Research)

Sentiment Analysis Performance of Lora-Enhanced Llm's

Tunahan TİMUÇİN

On social media platforms, the increase in the number of users and the resulting increase in thedata produced by users have accelerated the prominence of some technologies. The most well-known of these areas is Natural Language Processing (NLP), where sentiment analysis of userco…

Also available via: European Organization for Nuclear Research

View free PDFSource page
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-23

Reading creativity from the inside: a competing-routes view of the diversity–coherence trade-off in language models

Tomas Pødenphant Lund

Raising the temperature does not make a language model more creative, and this paper shows why at the level of the token race: temperature collapses value and coherence without ever raising novelty, while prompts that change which continuations can win raise judged novelty monoto…

View free PDFSource page
openalexZenodo (CERN European Organization for Nuclear Research)2026-07-26

Output Management Plan For Its Honesty and Integrity

K. Yamada

July 25, 2026. Paper Permission: What Publisher AI Policies Allow and Practice Withholds AFS3.6: Paper Permission: What Publisher AI Policies Allow and Practice Withholds Names Paper Permission: a permission present in the text of publisher policy and largely absent from open pra…

View free PDFSource page