arXiv_CL
-
C$^2$U in RAG: Compositional Concept Unlearning in Retrieval-Augmented Generation
2026-10-21 05:30:06Yiyang Duan, Fan Wu, Tiantong Wu, Che Wang, Zhiren Gong, Wei Guo, Yang Cao, Wei Yang Bryan LimAbstract
Retrieval-augmented generation can expose unwanted concepts through retrieved evidence even when the generator is moderated. We study query-side concept suppression with a fixed encoder and document index. Compositional Concept Unlearning (C$^2$U) routes a profile-derived direction through an augmented atom bank, adding only a bounded edited-minus-original delta to the raw query. Across five datasets, four encoders, and three seeds, the validation-selected configuration reduces pooled binary target Hit@5 by 4.25 percentage points with non-positive control degradation and target-exposure changes. Under the evaluated query transforms, it suppresses target density without increasing budgeted control-query exposure. Separately, ArXiv four-concept composition improves over an averaged-direction comparator after independent validation gates and first-use test evaluation.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29223
PDF
-
When Multiple Choices Are Correct: Asymmetric Probability Allocation in Jev and Other Decision Models
Abstract
Decision models return normalized probability distributions over candidate choices, which downstream applications can use to select among choices. Although their providers emphasize the reliability of these probabilities, calibration on tasks with a single correct choice does not establish their reliability when multiple choices are correct. We evaluate three decision models (Jev, Liquid d1, and GPT-6 Luna Decisions) on three diverse datasets (MetaTool, HotpotQA, and ECtHR) to study probability allocation and thresholding. Even when models rank every correct choice above every distractor, in 28% to 50% of those lists, at least one correct choice receives a probability below the uniform probability of 1/n, where n is the number of choices. We also present each model with five identical copies of a correct choice and observe probability disparities, with mean gaps ranging from 17 to 79 percentage points. We further show how these allocation disparities affect the recovery of multiple correct choices under probability thresholding. Decision-model probabilities can rank correct choices, but a probability threshold does not identify the set of correct choices.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29217
PDF
-
Preference Is Not Permission: Personalizing the Auto-Purchase Line in Agentic Commerce
2026-10-21 05:15:31Ryotaro Shimizu, Sai HtaungKham, Mo FanAbstract
Personalized shopping benchmarks ask which item fits the user; we ask how far an agent may commit on that item, and pose learning the individual's auto-purchase line as a task. The boundary is evidence-conditioned and three-valued: whether committing is appropriate depends on the sufficiency of product information (warrant), reversibility, price against the user's norm, and urgency, not only on fit; and the natural output is don't buy, confirm first, or buy automatically. Across sixteen instruction-tuned models, autonomous commitment is misaligned with a human authorization reference in which fifteen annotators, role-playing the delegating principal from a given profile, choose the furthest action they would allow without confirmation: models commit even when fit is low and when their own evidence flag reports the warrant missing. This is overreach. The line is individual: these raters disagree, yet the disagreement is structured and partly predictable from a person's own other labels. A shallow learner on a few dozen of a person's own labels predicts their held-out auto-purchase line at a median of 0.79 by 30-40 labels, outperforming every evaluated prompt$\times$model configuration, including a per-annotator oracle over 80 configurations; the learned boundary does not transfer across domains. We release the counterfactual benchmark, human authorization reference, and annotation protocol after acceptance.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29216
PDF
-
Understanding Target Probability Reuse in Speculative Decoding
2026-10-21 05:02:50Jungseob Lee, Giyoon Jang, Chailin Nam, Jiwoo Kum, Sihyeon Lee, Kyojun Choo, Jihun Song, Eunseo Song, Chanjun ParkAbstract
Speculative decoding uses a target model to verify a tree of draft tokens in parallel. We study whether the resulting next-token distributions can help select the next tree. For unrestricted target models, we show that guaranteed exact reuse requires a distribution computed at the next round's complete context. We characterize when retaining this distribution requires sacrificing accepted tokens and establish conditions under which this loss persists across decoding rounds. Across six targets, we find that reranking first-token branches with a separately computed distribution at the current context improves expected acceptance over draft ranking, especially with small tree budgets. Drawing the first token from this distribution before allocating the tree further improves expected acceptance over pruning a preselected tree, even at equal mean tree size. In our experiments, both selection rules also decode faster than draft ranking when the separate target pass overlaps standalone drafting. Our findings show that making target probabilities available before tree allocation can improve the efficiency of speculative decoding. Our code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29210
PDF
-
Adaptive Steering and Remasking for Safe Generation in Diffusion Language Models
2026-10-21 04:56:49Yejin Lee, Ungsik Kim, Yo-Sub HanAbstract
Diffusion Language Models(DLMs) provide a promising alternative to autoregressive language models through iterative denoising and bidirectional generation. However, their iterative generation process introduces distinct safety vulnerabilities because harmful content can emerge at arbitrary positions and persist across subsequent denoising steps. Existing defenses rely on fixed interventions or aggressive remasking, which limits adaptive control over denoising trajectories and can degrade generation quality. We propose an inference-time defense framework that combines adaptive safety steering with safety-aware remasking. Our method uses a gating direction to continuously adjust steering strength from the current denoising state and applies a steering direction to masked positions to guide subsequent predictions toward safer trajectories. Our method further employs a lightweight response detector after the first generation block to identify unsafe trajectories at an early stage. The detector triggers targeted remasking over generated content and part of the conditioning prompt, and the model regenerates the selected positions under adaptive safety steering. This design combines continuous trajectory control with explicit correction of unsafe content while requiring no modification of model parameters. Experiments on LLaDA and Dream demonstrate that our method improves robustness against diverse jailbreak attacks while preserving benign generation quality and general model capability. Our code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2605.13043
PDF
-
norregel: a preregistered benchmark of dated numeric values in Norwegian law
Abstract
norregel asks language models for numeric values in Norwegian statutes, regulations and parliamentary decisions (NOK amounts, percentages, multiples of the National Insurance basic amount (G) and age limits), each at a stated date. The 916 items were extracted mechanically from Lovdata's open data, and each carries the dated history of its provision, so an answer can be classed as correct, outdated, wrong or uninterpretable. We report a first and a second preregistered run on the same items. In the first run (five models, listed publicly only after local analysis), H1, that outdated answers are more frequent for values that took effect after a model's cutoff, could not be tested in substance, and H2 and H3 were inconclusive; Opus 5.5 and Sonnet 4.6 answered 46.7% and 40.2% of the 880 primary items correctly. The second run was preregistered and published before any of its calls, and eight models were analysed. Its pooled primary test supported H1 under the preregistered decision rule (d = 17.8 percentage points, 95% interval [7.7, 28.9], p = 0.0001). In the preregistered sensitivity analysis S8, which counts as outdated only the most recent value that differs from the key, d was 3.9 [-3.6, 12.7] (p = 0.1754), an interval that contains 0. H1 was supported in every other preregistered sensitivity analysis, but S8 is the only one that gives both groups one target value, so we do not regard the support for H1 as robust. The share of outdated answers equal to that most recent value (H2) was 56.1% [46.6, 65.0], an interval that contains 0.5, and the pooled test of a lower share correct in Nynorsk than in Bokmål (H3) was inconclusive. Several models, among them Haiku 4.5, gave mostly uninterpretable answers. Items and keys are withheld; item IDs, per-answer classes, code and results are on Zenodo.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29207
PDF
-
Reasoning-Search-Augmented LLMs: An Operational Taxonomy, Benchmark Analysis and Controlled Evidence Synthesis
2026-10-21 04:19:06Biswas Poudel, Nilson Chapagain, Amit Kumar, Xianshun JiangAbstract
Large language models (LLMs) demonstrate impressive emergent reasoning capabilities, yet they remain constrained by fixed model parameters and finite context windows. Even when augmented with static retrieval, they can produce statements that are unsupported by, or even contradictory to, retrieved evidence. This leads to hallucinations, difficulty incorporating up-to-date information, and failures to correctly interpret or reconcile sources. Recent work addresses these limitations by tightly coupling step-by-step reasoning with interactive search, treating retrieval as an action within an iterative agentic loop rather than a one-shot preprocessing step. This study reviews frameworks that implement such reasoning-search loops by interleaving planning, querying, retrieval, integration, verification, and stopping. We analyze these systems through four architectural axes: retrieval coupling, controller design, knowledge source, and grounding strictness. We separately treat the learning signal as a cross-cutting dimension because prompting, supervision, preference optimization, and reinforcement learning can instantiate the same architecture while producing materially different search, stopping, and grounding behavior. We further distinguish benchmark difficulty from RSA-loop demand and synthesize controlled within-study ablations and quality-cost deltas, avoiding invalid pooling across heterogeneous model backbones, retrievers, corpora, and evaluation protocols. Our goal is to clarify the landscape of reasoning-search-augmented LLMs and outline a unified perspective on the design choices shaping future research.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29189
PDF
-
LTEM: LLM-Friendly Table-Enhanced Markdown for Table Understanding and Generation
Abstract
Tables in real-world documents, including scientific papers, financial reports, and statistical summaries, frequently use merged cells, multi-level headers, and sectioned layouts. Yet the lightweight formats LLMs handle best (Markdown, CSV) cannot express these structures, while expressive formats (HTML, LaTeX) impose excessive token cost and parsing complexity. To enable systematic analysis of this trade-off, we introduce a taxonomy decomposing table structure into six atomic features (F1-F6) and corresponding complexity levels (L1-L6), and construct STB (Synthetic Table Benchmark), a controlled benchmark with 1,089 tables systematically distributed across complexity levels. Using this framework, we propose LTEM (LLM-friendly Table-Enhanced Markdown), a formally specified, minimal extension of Markdown that explicitly encodes spans and hierarchical headers. Experiments across five state-of-the-art LLMs show that LTEM achieves accuracy comparable to HTML, offers a favorable accuracy-token trade-off relative to HTML and LaTeX, and produces substantially fewer invalid responses than LaTeX; for table generation, LTEM is comparable to MultiMarkdown though HTML is best. We release our taxonomy, benchmark, and format specification to support future research on LLM-table interaction.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29188
PDF
-
TrainBridge: Agentic NPU Training Optimization Guided by GPU Compilation Evidence
2026-10-21 04:00:48Yupeng Chen, Kunxiang Zhao, Xingjian Wang, Yue Zhao, Yijing Wu, Ye Liu, Kai Li, Congliang Chen, Ruoyu SunAbstract
We study how to improve training efficiency on Ascend NPUs. In the NPU training programs we study, framework-level compilation optimizations such as operator fusion are only partially applied, leaving fragmented computation and avoidable execution overheads. Profiling-guided tuning can address these inefficiencies, but it often requires costly trial and error. To make training on NPUs more efficient, we propose TrainBridge, an agentic approach for transferring GPU optimization guidance to NPU training. TrainBridge uses the highly optimized execution produced by mature GPU compilation stacks as a task-specific reference for NPU tuning. It extracts optimization intents from the GPU training execution to guide NPU program rewrites and further refines the rewritten program through iterative profiling. Our GPU-to-NPU migration experiments show that TrainBridge achieves an average 1.7x speedup across four workloads covering LLM pretraining, image classification, and vision-language training, and an average speedup of nearly 2.0x across 27 Modded-NanoGPT speedrun records with highly optimized GPU implementations.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29171
PDF
-
TeachBack: Learning to Teach via Bidirectional Preference Optimization for On-Policy Self-Distillation
2026-10-21 03:59:23Linfei Ji (Huazhong University of Science and Technology), Jiayun Li (Tsinghua University), Haozhao Wang (Huazhong University of Science and Technology), Siyu Jiang (Huazhong University of Science and Technology), Zehao Wang (Harbin Institute of Technology), Longtao Xiao (Huazhong University of Science and Technology), Yu-Jie Yuan (Huawei Technologies Ltd), Ruixuan Li (Huazhong University of Science and Technology)Abstract
On-policy self-distillation (OPSD) enables a language model to learn from a Teacher instantiated from the same model but given access to privileged information, such as a reference solution. We find that access to sufficient privileged information allows the Teacher to generate highly accurate trajectories; however, accuracy alone does not guarantee that these trajectories are suitable for Student learning. Direct access to a complete solution may encourage shortened, reference-conditioned derivations that omit the exploration, verification, and self-correction required to solve problems without privileged information. We introduce TeachBack, a bidirectional preference-learning framework for on-policy self-distillation that learns not only from the Teacher but also how the Teacher should teach. TeachBack shapes the Teacher's generation behavior by controlling the length of the reference-reasoning suffix injected into the prompt and by varying the prompt style. While maintaining comparable Teacher accuracy, it encourages the Teacher to explore and re-solve problems, thereby reducing its reliance on the reference solution. At each rollout refresh, the shared policy generates both a privileged Teacher trajectory and a problem-only Student trajectory. A pairwise preference objective uses privileged Teacher trajectories to improve Student learning, while the reversed preference uses the Student's behavior to shape future Teacher supervision under the privileged condition. Experiments on mathematical and logical reasoning tasks show that TeachBack consistently improves Student performance over existing OPSD methods and related baselines.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29167
PDF
-
Why is AI-generated language so hard to understand?
2026-10-21 03:54:08Thomas Hikaru Clark, Matias B. Martin, Kyle MahowaldAbstract
There is a widespread sense that AI-generated text has recently become harder to understand. In this study, we investigate a set of psycholinguistic features that are known to correlate with human processing difficulty (e.g., object relative clauses, long syntactic dependencies, low-frequency n-grams). Using a novel corpus of AI-generated text across diverse domains, we find that AI models use these difficult features more often than paraphrases generated by round-trip translation. Compared with human writing, all nine frontier and near-frontier models use more inanimate subjects with animate verbs and more low-frequency n-grams, and most also use longer dependencies, more stacked nouns, and more reduced object relatives in conversation; we also demonstrate substantial variation across model families. We propose and test a hypothesis as to why: while human language is optimized for successful communication, AI-generated text is increasingly optimized for other goals like math, code, and reasoning, in which there is a pressure for \textit{information density} rather than successful communication. We provide a proof-of-concept for this hypothesis by fine-tuning a language model under a length penalty, which leads to a divergence from human-like linguistic behavior. Our findings highlight the axes of misalignment between linguistic processing demands in humans and LLMs, and point toward possible remedies.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29162
PDF
-
VeeReview: Measuring Build-Verify Pairs in Agent-Authored Pull Requests from a V-Model Perspective
2026-10-21 03:25:44Gabriel Dautovi, Jie JW WuAbstract
Context: Reviewing a pull request requires verification evidence that addresses the proposed change. Coding agent benchmarks and studies of pull request outcomes do not directly assess this relationship. Objective: We test whether the V-Model's build and verify pairs can measure verification evidence in pull requests and support comparisons between agent-authored and human-authored changes. Method: VeeReview classifies four build and verify pairs using repository signals and a language-model judge. Three pairs use model judgements constrained by deterministic signals, and a rule over the diff and repository tree decides Module Design-Unit Test. We evaluated VeeReview against a 25-pull request reference set labelled by two raters and applied it to 1,454 matched pairs of agent and human pull requests from 741 repositories. Results: VeeReview assigned lower build-only rates to agent-authored pull requests on all four pairs, contrary to the preregistered hypothesis (mixed-effects odds ratios 0.52 to 0.86). Architecture-Integration was not significant. Standardization for linked issues, shipped tests and size reduced matched-rate differences by 34% to 60%, and the rule-decided pair's standardized interval included zero. The instrument did not meet its preregistered calibration criterion: agreement with adjudicated labels was 4% to 60% against a required 80%, and inter-rater agreement was weak (kappa = 0.154). Conclusions: The rule-decided pair shows that a process measure judged by a language model partly reflects how a pull request is presented rather than the verification work behind it. We report the cohort differences as observations made with this instrument, and we argue that such measures need a mechanically decided component and a validated human reference before they can support comparisons of verification practice.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29143
PDF
-
GRAFT: Restoring the Missing Sampling Interface in Continuous Diffusion Language Models
2026-10-21 03:15:30Haoning Yang, Lili PanAbstract
Continuous diffusion language models generate a whole sequence in parallel, never drawing a single word at any intermediate step; only the final step turns embeddings into text. Models such as ELF now report autoregressive-level quality at a fraction of the sampling steps of discrete-token diffusion models, and this design principle -- no step-wise discretization -- is what their own ablations credit for the gain. We show that this same principle carries a hidden cost: a collapse of the lexical long tail. Measured against held-out human text, ELF's output matches the human rate on common words almost exactly, yet its rarest words appear at only 0.30x the human rate, a gap that does not close from 105M to 652M parameters and that no sampler setting these models expose can reach. A same-size GPT-2, in contrast, spans zero to above the human rate by its decoding rule alone. We trace the difference to what continuous diffusion models lack: with no step that ever draws a discrete word, every step is instead driven by an average over the candidates still in play, and averaging a rare candidate together with the common word competing for its slot lands on the common word. An autoregressive model avoids this because its decoding rule samples one candidate rather than averaging over all of them. We propose GRAFT, a training-free method that grafts this missing choice back onto a frozen model: while the model is still undecided, GRAFT commits the rare words the average would otherwise discard, then regenerates the sequence from scratch with those commitments held in place. Across three ELF scales and on LangFlow, an independent continuous dLLM, GRAFT restores the rare-word share of generated text from 0.42-0.77x the human rate to 0.82-1.10x, a relative gain of 34-109%, and moves the entire word-frequency profile closer to human on every model tested, by 2.7x on ELF-B and 12x on LangFlow.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29136
PDF
-
GRAM: Group-wise Rank-Aware Model Merging via Subspace Alignment
Abstract
Merging different models into a single model is highly desirable for multi-task deployment, particularly in regulated domains like healthcare. In such settings, models are often locally fine-tuned due to strict data privacy policies, necessitating that aggregation operates exclusively on the fine-tuned weights. Existing methods apply a uniform merge across all tasks, ignoring the geometric relationships among their learned subspaces. When task updates occupy misaligned subspaces, projecting them onto a shared low-rank basis discards task-relevant directions, a failure mode named \emph{subspace interference}. Through this study, we show that the per-task projection error of a merged model is governed by the principal-angle alignment between each task and the shared subspace, so that grouping geometrically compatible tasks provably reduces this error within each group. Motivated by this observation, we propose \textbf{Group-wise Rank-Aware Model Merging (GRAM)}, which uses the principal-angle geometry of task vectors to decide which tasks should be merged together and then performs a standard SVD-based merge within each coalition; the pipeline requires only the fine-tuned weights at merge time. With only one additional model ($\nK{=}2$), GRAM significantly mitigates subspace interference and consistently outperforms prior merging methods across language understanding (GLUE with Flan-T5), 8-task CLIP vision, and medical instruction tuning (Qwen2-7B-Instruct with four English/Chinese medical LoRAs).
Abstract (translated)
URL
https://arxiv.org/abs/2610.29110
PDF
-
Multi-level Encoder with Global Topic for Task-oriented Dialogue Summarization
2026-10-21 02:41:38Zhuoqi He, Peijie Huang, Yuhong Xu, Youming Peng, Mingzhi Xu, Xinyang LinAbstract
Task-oriented dialogue summarization aims to automatically extract key information to generate domain summaries to improve service efficiency and quality. Task-oriented dialogue is inherently logical and surrounds specific topic. How to effectively capture the dialogue topic and the most salient information becomes one of the major challenges of this task. In this paper, we propose a task-oriented dialogue summarization model utilizing a multi-level encoder with global topic (MLEGT). This model discovers temporal and spatial dependencies through a multi-level encoder, while simultaneously obtaining a simple yet effective global topic in the process of graph construction. It not only captures the crucial details but also breaks the long-range limit and discover the intrinsic structure across the dialogue. Then, a global topic guided pointer mechanism is added into the summrizer to capture ignored but important information, which is helpful to improve the accuracy of the summary. A comprehensive study of two public datasets, including a real-world large-scale Chinese police interrogation dataset and a medical reporting service dataset proves the superiority of our method on several strong baselines.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29097
PDF
-
Same Answer, Different Reasons: Trace-Level Divergence in Multilingual Large Language Models
2026-10-21 02:35:06Rui Sun, Peizhao Mei, Xiewei Cheng, Kaizhuo Chen, Taoran Li, Zhiyuan Yu, Varun ChandrasekaranAbstract
Multilingual large language models are increasingly deployed in settings where visible reasoning traces support verification, education, and safety auditing, yet standard evaluations typically measure only final-answer agreement or holistic response similarity. As a result, they establish outcome consistency without determining whether users across languages receive comparable supporting evidence. We introduce a trace-level divergence framework that compares visible reasoning traces as structured sequences, decomposing agreement into structural similarity, order consistency, and holistic similarity, with pair-specific calibration to enable comparison across languages. Across three reasoning benchmarks, seven languages, and six frontier LLMs, the natural baseline---embedding each whole response and comparing the two by cosine---systematically overstates agreement: $52.1\%$--$67.3\%$ of examples in the top quartile of holistic similarity are not in the top quartile of structural similarity. This dissociation is stable across sentence encoders and step segmentations, and blind bilingual annotation studies in two English-anchored language pairs show that human judgments align with structural similarity rather than with the holistic baseline. A same-language control shows that the level of divergence is language-specific: two English traces for paraphrases of the same question agree more closely than an English trace and its counterparts in other languages. Our results show that answer-level and holistic evaluations can substantially overestimate the consistency of multilingual explanatory artifacts, motivating trace-level evaluation for reliable multilingual deployment.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29092
PDF
-
Copying Under Repetition: From Prefix Matching to Position Hashing
2026-10-21 02:21:26Anxin Guo, Tianlong Huang, Chenxiao Yang, Zhiyuan LiAbstract
Frontier LLMs can copy random text almost flawlessly, yet they often overrun or truncate long blocks of repeated characters. In repeated strings, matching local content no longer reliably identifies where copying should continue or when it should stop. Nevertheless, we show that a two-layer transformer with $O(\log\log N)$-bit precision and $O(\log N/\log\log N)$ width can copy every string up to a chosen design horizon $N$ and terminate correctly under arbitrary repetition. We use RoPE-style rotations of attention values and outputs (VO-rotations) with small prime periods to encode distances from delimiter anchors modulo those primes. By the Chinese Remainder Theorem, these periodic encodings jointly distinguish all required distances, allowing one retrieval rule to select either the next source token or the end marker. In contrast, we prove a width-independent impossibility result for copying under sequential finite-precision softmax accumulation. It applies to transformers whose only positional input is an additive attention bias that eventually becomes constant, including NoPE and a recursive implementation of ALiBi. Once an identical-token run exceeds the resulting saturation threshold, further repetitions leave the logits unchanged. If this threshold grows sublinearly with $N$, these models cannot both complete a long all-identical copy and stop correctly, a pattern resembling the termination errors of frontier LLMs. In synthetic copying experiments, baseline transformer models fail on long repeated strings even within the training range. Trained VO models, by contrast, generalize beyond the training lengths with near-perfect accuracy up to length $\approx5005$, the selected primes' period product.
Abstract (translated)
URL
https://arxiv.org/abs/2610.26722
PDF
-
A Large-Scale Evaluation of Role-Enhanced Zero-Shot Chain-of-Thought Prompting
2026-10-21 01:53:34Ibrahim Khan, Thai Van Nguyen, Yi Xia, Pratch Suntichaikul, Mustafa Can Gursesli, Daniel Fernandez Galeote, Juho Hamari, Ruck ThawonmasAbstract
This paper proposes Role-Enhanced Zero-Shot Chain-of-Thought (RE-ZS-CoT), a prompting framework that systematically integrates domain-specific expert roles to enhance reasoning in large language models (LLMs) without task-specific examples. We assess two versions of our approach. In the first version, static role selection, we evaluate eight predefined expert roles across arithmetic, commonsense, and reading comprehension benchmarks to determine their impact on reasoning performance. In the second version, dynamic role selection, each model autonomously selects the single most suitable role for each benchmark, enabling adaptive task-role alignment without manual specification. Across 20 LLMs and seven benchmarks, RE-ZS-CoT frequently surpasses zero-shot and zero-shot-CoT baselines, achieving performance improvements of up to 44 percentage points. The results reveal a strong correspondence between static and dynamic versions; the benchmarks where static roles yielded improvements were almost always those where dynamic roles also outperformed baselines, though static selection achieved peak performance. Ablation studies further show that unconventional roles degrade performance, underscoring the necessity of task-role alignment. The framework's scalability is validated across model sizes (2b-14b parameters), with smaller models like Granite3.3-8b achieving remarkable proficiency in arithmetic reasoning. RE-ZS-CoT advances zero-shot prompting by unifying role specialization with chain-of-thought reasoning, offering a practical, resource-efficient solution for improving LLM capabilities across diverse domains.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29073
PDF
-
Our Own Jev: Typed Decisions in One Forward Pass, from an In-House Judge to Open LoRA Models
2026-10-21 01:28:58Yuchuan Weng, Chi Liu, Xin Chen, Srinivasan ManoharanAbstract
A typed decision maps evidence, a question and a fixed set of options to a probability for every option, computed in one forward pass, by reading a language model's logits or a small head over its hidden states, instead of generating and parsing text. The commercial system Jev popularised this System~1 interface, and open reproductions have followed. We report two uses of the pattern. \emph{Part I} describes an in-house System~1 judge that triages about 148K agentic coding session records for a training corpus. It replaces a frontier LLM judge that cost on the order of \$1,000 per pass and whose stored labels matched a fresh two-judge consensus on only 26.9\% of the sessions where the two judges agreed; a self-hosted frozen backbone with a rank-16 LoRA adapter and a pointer head answers three typed questions in one forward pass, judges 1,000 sessions in about three minutes of one GPU, about \$0.06 at a reference rate and roughly 1\% of the frontier judge's cost, and keeps every session inside the organisation. \emph{Part II} asks whether a readout that adds no parameters can be competitive and whether training one helps. Applying the public Eikos recipe and corpus of about 25K decisions to three further backbones (1B, 2B and 35B), we read the frozen output layer of a LoRA-adapted decoder. On the 111 public hard items of JevBench our single-seed 35B, developed with those items in view, scores 79.3\%, not significantly different from the commercial reference (73.0\%, unadjusted $p=0.12$) and ahead of five open systems after Holm--Bonferroni correction; our 2B is ahead of all four kev checkpoints on the leaderboard. A single-seed ablation finds no detectable benefit from training the 26 letter rows of the output layer, and a retrain on a later corpus release moves the 2B's hard-tier accuracy by 11.7 points, which shows how little small evaluations resolve.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29052
PDF
-
HybridResearcher: Bridging the Structured-Unstructured Boundary for Deep Research Agents
2026-10-21 00:41:07Soyoung Yoon, Peiran Xu, Han Wang, Ruofan Wu, Xiaodong Yu, Feng Yan, Mohit Bansal, Debiao Li, Seung-Won Hwang, Yuxiong He, Zhewei YaoAbstract
While LLM agents are increasingly capable of gathering and synthesizing evidence from a single homogeneous source, such as structured tables or unstructured data, real-world analytical questions frequently span heterogeneous sources. For example, answering one enterprise question may require both a result computed over an internal database and evidence found on the web. Recent evaluations on benchmarks like HybridDeepResearch show that even frontier agents struggle when evidence must cross this boundary. While these benchmarks have exposed the limitation, methodologies to explicitly teach cross-source reasoning remain absent. To address this, we introduce HybridResearcher, a 4B agent trained to bridge evidence across the structured and unstructured boundary. To build it, we synthesize cross-source bridge trajectories by starting with validated single-source tasks-such as existing text-to-SQL or web search queries-and fabricating the missing source around a bridge entity drawn from the seed's own output. We retain a sample only when strict construction gates confirm that neither side alone suffices, ensuring the data requires true cross-source interdependence. Finally, we train on this curated dataset using supervised fine-tuning followed by reinforcement learning. Evaluated on the existing HybridDeepResearch benchmark, HybridResearcher improves over our base model, Qwen3.5-4B, by 30.0 points pass@8 and surpasses the far larger Qwen3.5-397B-A17B by 8.3 points, while also improving over its base model on established, held-out single-source benchmarks (BrowseComp and BrowseComp-Plus), demonstrating that this complex cross-source reasoning can be taught directly to small models without triggering catastrophic forgetting.
Abstract (translated)
URL
https://arxiv.org/abs/2610.24899
PDF
