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.
By Gnaneswar Villuri, Hashmath Shaik, Alex Doboli
arXiv:2607. 26102v1 Announce Type: cross Abstract: Mathematical chain of thought (CoT) evaluation is commonly reduced to whether the final answer matches a reference.
By Vivek Shukla, Varun Shukla, Atul, Divya Mishra, Mehul Kumar Das
arXiv:2607. 14552v1 Announce Type: cross Abstract: A standard recipe for distilling the reasoning ability of large language models (LLMs) is to sample chains of thought from the model, keep those that reach the correct final answer, and fine-tune on the survivors.
By Jungseob Lee, Seungyoon Lee, Suhyune Son, Dongyub Jude Lee, Sungbin Han, Sugyeong Eo, Heuiseok Lim
PROOF is a benchmark that profiles the reliability of object-level facts in instruction-tuned language models by converting a frozen Wikidata snapshot into 18,486 English multiple-choice questions grounded in 11,779 semantic facts across 101 classes, 392 properties, and 14 domains. Each question includes an explicit "I don't know" option, a "No correct option" control, and nine controlled formulations, with 1,849 questions designed as no-correct-option traps. The study evaluates 18 open-weight model deployments on 166,374 prompts, revealing wide variability in factual accuracy, sensitivity to wording changes, and the impact of decoder perturbations.
By Andrei Chetvergov, Mikhail Solovev, Timofei Sivoraksha, Stepan Ukolov, Valeriia Kuschenko, Alexander Evseev, Sergey Bolovtsov
The study shows that large language model (LLM) agents are far more likely to commit to a directional prediction when presented with a professional‑looking market panel than when asked the same question directly, with commitment rates rising from 6.5% to 54.0% across 12 frontier models. Even when the panel’s data is entirely fabricated, commitment still increases significantly, indicating that the authority of the presentation, rather than the truth of the information, drives confident action. The authors demonstrate that this act/don’t‑act decision gate is narrow, model‑specific, and can be mitigated through supervised fine‑tuning, though its effectiveness depends on response format and context.
whyItMatters":"The findings reveal a specific vulnerability in LLMs where presentation style can override factual accuracy, highlighting the need for careful design and training to prevent misleading confidence in uncertain scenarios."
By Pranav Aggarwal
The paper investigates whether a language model’s own confidence can replace labeled data for teaching it to abstain from uncertain answers. By fine‑tuning models with LoRA to answer only when their frozen confidence is high and to say “I’m not sure” otherwise, the authors show that this label‑free approach matches label‑supervised abstention tuning on short‑form factual QA. The method works across six open‑weight models (1B‑8B) and is effective except for confidently wrong facts, which the confidence signal cannot flag.
By Ali Asaria, Tony Salomone, Deep Gandhi
arXiv:2608.22483v1 Announce Type: new
Abstract: Large Language Models (LLMs) increasingly support decision-making in high-stakes domains, but they often hallucinate and express confidence that is mis...
By Toghrul Abbasli, Kentaroh Toyoda, Yuan Wang, Li Chen
arXiv:2608. 15445v1 Announce Type: new Abstract: When a reward is correct on every training example yet consistent with more than one goal, a model can acquire an unintended one, a failure known as goal misgeneralization.
By Suyash Maniyar, Armaan Sandhu, Abhishek Mishra
arXiv:2609.05928v1 Announce Type: cross
Abstract: Large language models now compute correct tax liabilities on over 90% of well-formed cases in statutory benchmarks, which makes them candidates for t...
By Albert Sadowski, Jaros{\l}aw A. Chudziak
The paper investigates whether giving AI monitors access to the final answer improves their ability to verify reasoning. Using 237 step‑by‑step solutions to physics exam questions, the authors found that answer access mainly helps monitors detect inconsistencies with the final answer rather than independently checking the reasoning. Certification of the answer increased overall accuracy and error localization but reduced the ability to flag critical traces where the answer was correct but the reasoning was flawed.
By Will Yeadon, Sergio Ju\'arez, Paul Mackay, T. J. Dowling, Elise Agra, Oto-obong Inyang, Arin Mizouri, Craig P. Testrow
The paper introduces a diagnostic for reference‑free judge gates in text‑space skill optimization. It formalizes a judge as a latent solver, deriving a closed‑form bound on discriminability (ROC‑AUC) in terms of judge competence and answer‑space size, and shows that discriminability is confounded by item difficulty unless a within‑question estimator is used. A non‑intervening probe demonstrates that discriminability is at chance near the competence floor, rises above it, and that the diagnostic can predict gating errors in closed‑loop experiments.
By Chenle Chen, Yangbo Wei, Chao Yao, Shaoqiang Lu, Junhong Qian, Chen Wu, Lei He
The paper shows that large language model agents are far more likely to commit to a directional answer when presented with a professional-looking market panel, even if the panel’s data is fabricated. Across 12 frontier models, commitment rates jump from 6.5 % for a bare question to 54.0 % with evidence, and remain high (≈37 %) even when all numbers are invented. The study finds that the act/don’t‑act decision gate is the key failure point, can be trained to reduce false commitments, but is fragile to response format changes.