Shortcut Before Circuit: Document Statistics Time In-Context Conflict Resolution

Detectable Only Where It Is Confounded: What Verified Duplication Counts Say About Membership Evidence in Language Models

When a language model finds a sentence unusually cheap to predict, it is tempting to conclude that the sentence was in its training data. Almost every published test of that inference has had to guess which sentences were in the training data, the members, and which were not. This paper removes the guessing. Two model families, OLMo-2 and Pythia, publish their pretraining corpora, and a public index over those corpora returns the exact number of times any sentence appeared in each. Those counts make three questions answerable directly. The answers form a pincer, closing from two sides. At the duplication levels ordinary text actually has, five models from 1B to 13B parameters carry at most a faint trace of their own exposure. We measure that trace with a design that reads the same sentence through two models, which cancels fluency and quality by construction, and it comes to a rank correlation near -0.08, where -1 would be a perfect relation and 0 none. Where the trace does become strong, above roughly a thousand copies, the two corpora agree on which sentences those are, because they are the famous ones, so exposure can no longer be told apart from fame. Two further measurements show how apparent membership signal gets manufactured. A common way to build a non-member is to change one word of a member. The model does prefer the original, but the gap is the same whether the original appeared once or a hundred times, so what the model is rewarding is the author's word choice, not memory. Above a thousand copies the gap grows with model size on the twelve sentences we can test there, at the same boundary where the pincer closes. And swapping the controls for sentences that differ from the members in register moves a detector from 0.83 to 0.94 AUC, on a scale where 0.5 is a coin flip and 1.0 is perfect separation. We release the sentence banks, counts, and code.

Where and When to Commit: Candidate-Aware Decoding for Diffusion Language Models

Actions Speak Louder than Words: Measuring Cross-Lingual Policy Retention in Tool-Using Agents

A single trace-extraction regex, not the model, manufactured a multilingual failure: two worked examples raise one model's measured accuracy twenty-sixfold while its accuracy on readable outputs barely moves.

Explicit, Not Longer: What Makes Epistemic Stance Survive Memory Compression

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. We ask what governs whether it does. Matched notes carry the identical claim and identical stance and differ only in where that stance sits; one model compresses both under the same budget among the same filler notes, and a blind reader that never sees the condition scores the result. Across 60 claims in seven registers, writing the stance as a labelled field rather than a bracketed aside raises retention by about 15 points on two models (37 claims to 2 on one, 30 to 8 on the other; permutation p=0.00005), and a pre-registered replication on Haiku, its prediction and decision rule committed before the run, gives +15.6 points, 38 claims to 1. Ablating the format on both models gives the same net effect from different parts: labels help on both (+9.7 and +12.8) and length helps on neither, but wording the stance as a full sentence is the largest component on one model (+12.5) and worth nothing on the other (+0.6). Either model alone would have licensed a confident and different mechanism, so we claim only the intersection: make the stance explicit, not merely longer, and expect the best way of being explicit to depend on the model. A deterministic readout with no model reproduces the two-cell direction and five of seven ablation contrasts, but not length or labels, which we therefore do not claim on one instrument. Fifty hand labels (kappa=0.75) agree on direction; we print their seven disagreements in full. We also report nine withdrawn claims, three of them former title claims of this paper.

Answer First, Reason Later: When Commitment Order Costs Accuracy in Diffusion Language Models

Masked diffusion language models revise many masked output positions in parallel. We call a token committed once it becomes visible and is never masked again, and call a response answer-first when the final answer commits before the reasoning printed ahead of it. On 1,069 GSM8K test questions, an explicit step-by-step instruction increases the accuracy difference between unrestricted decoding and a decoder that permits commitment only near the left-most unresolved position; unrestricted decoding also produces more answer-first trajectories. On MATH-500, the two LLaDA models spend most of a short output canvas on reasoning that commits after the answer, and the benefit of frontier gating decreases as that postanswer writing disappears. Dream-7B has little post-answer writing and follows a different accuracy pattern. A controlled four-option task reserves a one-token answer position before generation. Delaying that position outperforms an equally timed reasoning-token delay on LLaDA-8B, LLaDA-1.5, and Dream-7B. The raw difference is largest on Dream, whose free accuracy on the controlled task is lower. Answers commit much earlier under the reserved-position interface than in ordinary free-form generation, which limits how far the intervention result can be generalized. Commitment order affects the context used to complete a response and the allocation of a finite output canvas.

Attention-Path Fragility as an Uncertainty Signal in Large Language Models

It is proposed 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, a training-free estimator that masks attention heads and measures the BALD mutual information among the resulting subnetworks.