This study independently reproduces the dissociation reported by Zhao (2026) regarding chain-of-thought entropy in large language models. It confirms that the shape of the entropy trajectory predicts answer correctness, while the total entropy drop magnitude does not, across four open-weight models and two benchmarks (GSM8K and MATH‑500). The reproduction also maps settings where the magnitude signal holds or fails and documents protocol differences not reported in the original work.
This empirical study is an independent reproduction of the dissociation Zhao reported in 2026. The shape of a large language model's chain-of-thought entropy trajectory predicts whether the final answ...
arXiv:2607. 08059v1 Announce Type: cross Abstract: Uncertainty quantification for visual language models (VLMs) conventionally targets the answer token distribution.
arXiv:2604. 08941v2 Announce Type: replace Abstract: Medical Vision-Language Models (VLMs) answering binary presence questions on chest radiographs can fail in two linked ways: they are confidently wrong, and they change answers when a clinically equivalent question is rephrased.
arXiv:2609.07901v1 Announce Type: new Abstract: Weight quantization largely determines the economics of serving open-weight LLMs. Its costs are usually assessed with capability benchmarks, on which 4...
arXiv:2607. 18292v3 Announce Type: replace-cross Abstract: Bigger language models are less reliable.
arXiv:2609. 31181v1 Announce Type: new Abstract: Black-box model identification works by scoring a model's response to natural-language prompts.
A 0.6B language model consistently answers YES to 1,200 logical tests, yet its behavior shows no discrimination. Linear probes reveal the correct verdict with high AUC (0.96) and transfer to unseen structures, but a single scalar readout fails due to a saturated decision threshold offset by +4.6 σ. Adjusting this threshold restores behavior accuracy from 50 % to 81 % and improves higher‑scale models, demonstrating that miscalibrated readouts, not hidden knowledge loss, drive performance gaps.
The paper investigates how entropy over chain‑of‑thought tokens influences policy decisions such as gradient application, pruning, and collapse detection. By separating scaffold tokens from substantive content, the authors analyze entropy, Kullback–Leibler divergence, and entropy velocity for each channel, proving differences between raw and content conventions and bounding answer diversity. Empirical results across 23 configurations show that scaffold tokens can account for up to 41% of high‑entropy positions, with entropy share growing through distillation, while content conventions outperform raw surprisal on compression tasks and reveal significant answer leakage in re‑fed chains.
arXiv:2607. 07670v1 Announce Type: cross Abstract: Large language models hallucinate most about entities they have never seen.
The study measured the impact of a single training example on a GPT‑2 model by running 24 counterfactual experiments. 32 models were trained from scratch on OpenWebText, and at a specific training step a single batch row was replaced with a 194‑token passage under three conditions (fluent prose, fabricated subject, random characters) or left unchanged. Results showed that the passage was learned from one exposure and decayed, with measurable differences in cross‑entropy up to 50 steps after injection but no lasting effect at the final step.
The study examines how two small instruction‑tuned language models, Qwen2.5‑1.5B and Llama‑3.2‑1B, respond to user pushback on TriviaQA. When initially correct, the models flip to a wrong answer in about 42–43% of cases, with the effectiveness of different pushback styles varying by model. Attempts to decode capitulation from the pre‑response residual stream fail under a rigorous validation protocol, revealing overfitting and a measurement hazard that underestimates capitulation by 18–24 percentage points.
arXiv:2608. 03842v1 Announce Type: cross Abstract: When a language model fails on surface-perturbed input (typos, OCR noise, homophones), "which layer is responsible" has three natural operationalizations: where representations diverge most (sensitivity), where restoring clean activations recovers the prediction (causality), and where a small adapter can repair the damage (compensatory capacity) - and we show these three layer maps dissociate.