arXiv Computation and Language
Sep 28
Where a Model Sends Its Own Repeated Token
arXiv:2609. 31181v1 Announce Type: new Abstract: Black-box model identification works by scoring a model's response to natural-language prompts.
The Announcement Carries the Cue: Markup, Boundaries, and the Notation of Pre-Training Corpora
arXiv:2608. 09093v1 Announce Type: cross Abstract: How a document's arrangement is written down, its notation, is a training variable that no dataset card records.
Related stories
arXiv Machine Learning
3d ago
Forecasting Corpus-Sampling Disagreement in Masked Text Reconstruction
The paper investigates how much a masked‑text reconstruction predictor’s outputs would differ if a second corpus were collected from the same population. Using a smoothed predictor based on token‑pair counts, the authors derive a first‑order formula for expected disagreement between predictors trained on two independent corpora, accounting for document length, random context counts, and token‑offset correlations. Experiments on TinyStories, WikiText‑103, and CNN/DailyMail show that increasing corpus size from 512 to 2,048 chunks of 64 tokens reduces disagreement only modestly (5–40 %), and that forecast accuracy improves with larger reference corpora and a source‑split jackknife correction. whyItMatters:"The study provides a practical method to estimate how much predictor outputs would change with additional data, informing decisions about data collection and model robustness in masked‑text reconstruction tasks."
arXiv Machine Learning
Aug 4
Leak It: A Probabilistic Approach to Training-Data Extraction from Black-Box Language Models
arXiv:2608. 00144v1 Announce Type: new Abstract: Membership inference (MIA) on language models is usually summarised by an aggregate ROC-AUC, but such evaluations are confounded: model-free blind baselines separate members from non-members from surface text alone.
arXiv AI
Aug 24
Structure for Reading, Prose for Writing: Asymmetric Structural Conditioning in Multi-Agent Document Authoring
The paper reports on a deployed multi‑agent tender‑response system that uses an open‑weights language model under sovereignty constraints. In a blind comparison, the system’s answers were judged at least as good as human‑written bids in 40 of 55 sections, with only a few gaps attributable to missing knowledge rather than writing quality. The study also demonstrates an asymmetry in conditioning: while structural markup improves reading tasks, converting instruction material from prose to nested XML degrades answer quality, and naming forbidden constructions concentrates defects.
arXiv Machine Learning
Sep 11
Detectable Only Where It Is Confounded: What Verified Duplication Counts Say About Membership Evidence in Language Models
The paper investigates whether language models can identify sentences from their training data by using exact duplication counts from publicly released corpora for two model families, OLMo‑2 and Pythia. It finds that for typical duplication levels, models show only a weak trace of exposure, with a rank correlation near –0.08, and that strong signals only appear when a sentence appears roughly a thousand times, at which point fame rather than memory dominates. The study also demonstrates that common membership tests can be misleading, as changing a single word does not alter the model’s preference, and that controlling for register can significantly improve detector performance.
arXiv AI
Aug 17
Split the Labor: Separating Evidence Interpretation from Decision Aggregation
arXiv:2608. 14509v1 Announce Type: new Abstract: Systems that ask a language model to reach a conclusion from many sources usually concatenate them into one prompt.
arXiv AI
Aug 12
Attention-Path Fragility as an Uncertainty Signal in Large Language Models
arXiv:2608. 11138v1 Announce Type: cross Abstract: We propose that a model's uncertainty about a token is reflected not only in the breadth of its output distribution but also in whether a confident prediction is \emph{fragile} under perturbation of its attention pathways.
arXiv Machine Learning
Aug 12
Leak It: Per-Document Extraction Beyond Aggregate Membership Inference
arXiv:2608. 00144v2 Announce Type: replace Abstract: Membership inference (MIA) on language models is usually summarised by aggregate ROC-AUC, but such evaluations are confounded: model-free blind baselines can separate members from non-members using surface text alone.
arXiv AI
Oct 7
Isotropic Yet Undecodable: The Sequential Content-Sufficiency Gap in Latent-Predictive Text Representations
The paper investigates whether latent representations of text preserve the ordered information present in the input. Using an information‑theoretic decomposition, the authors show that even perfectly isotropic Gaussian representations can contain no target information, and that rank alone does not predict prediction risk. They introduce CANOPE, a non‑autoregressive framework with ordered latent canvases and geometric regularization, and demonstrate that while latent‑agreement and token‑grounded models achieve similar ranks, only the token‑grounded variant yields high positional recall and low word error rates on corrupted data.
arXiv AI
Oct 5
On the Tip of the Tongue: Why LLMs Hallucinate Answers They Can Decode
arXiv:2603.13911v2 Announce Type: replace Abstract: A language model can give the wrong answer even when the correct answer is decodable from its intermediate states. To study this gap between decoda...
arXiv Machine Learning
Aug 6
When More Becomes Less: Position-Dependent Repetition Effects in Language Models
arXiv:2608. 04021v1 Announce Type: cross Abstract: Cloze-style probes that vary how often a target token appears implicitly assume that more copies of a target affect prediction the same way regardless of where the readout slot sits.
arXiv AI
Aug 10
Explicit, Not Longer: What Makes Epistemic Stance Survive Memory Compression
arXiv:2608. 06953v1 Announce Type: cross Abstract: Agent memory systems compress what they store, and compression is built to drop qualifiers, so a claim's epistemic standing tends not to survive being written to memory.