Named after the hundred-eyed watchman of Greek myth, Argus watches the education landscape: spotting new opportunities, pressure-testing the ventures we're building, and tracing every read back to the real-world signals behind it.
The evidence library: the raw signals the pipeline is watching across the education ecosystem. Every idea is built from these.
arXiv:2608.25824v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used as evaluators to assess output quality and preference alignment, yet providing reliable guarantees of agreement with human judgments remains challenging. Recent work introduces confidence-thresholding methods that provide such guarantees for pairwise comparisons, relying on the assumption that higher estimated confidence implies lower disagreement risk with humans. However, this assumption can break down when the number of candidate responses increases, since distributing probability mass across many alternatives can distort confidence estimates. To address this issue, we propose a Localize-Then-Decide framework. First, conformal prediction localizes a small shortlist that contains the human-preferred response with high probability. Then, a calibrated confidence-based rule selectively chooses a single response from this shortlist or abstains. This design restores the monotonic relationshi
arXiv:2608.25768v1 Announce Type: new Abstract: Turkish encoder models have adopted modern architectures while leaving the pretraining objective fixed at masked language modelling. This paper introduces MoganBert-TR, a 149M-parameter Turkish encoder foundation model trained from scratch on a language-specifically filtered corpus, together with an embedding model derived from it (MoganBert-Embed). MoganBert-TR is trained over 237.3B tokens with a two-stage CLM-to-MLM curriculum: causal language modelling first, masked language modelling for the remainder, with the transition made inside the stable phase of a WSD schedule. In a controlled ablation under an equal step budget, this design outperforms pure MLM by 2.7-3.7x on Turkish MS MARCO retrieval; the measured mechanism is embedding geometry, where a single direction absorbs 28.1% of the variance under pure MLM against 11.9% under the curriculum. Long-context extension and learning-rate decay are then split into two branches after a sh
arXiv:2608.25761v1 Announce Type: new Abstract: One common trade-off in the use of large language models involves reducing the size of the model while increasing the amount of computation at inference time, for example by using a wider beam search. In this paper, we examine the constrained case of this "model size vs. inference compute" trade-off, in which the model outputs are constrained by a strict grammar at inference time. Our results demonstrate that the constrained trade-off behaves differently from the unconstrained trade-off. We investigate the task of converting a prose query into an equivalent SQL query (text-to-SQL). Performance is evaluated on the Spider text-to-SQL benchmark, using the Qwen2.5-Instruct model family ranging in size from 0.5B to 7B parameters, all at 4-bit precision. We experiment with two approaches to varying inference compute: (i) beam search with a variable number of beams; and (ii) sample+vote, i.e., sampling several constrained outputs and then voting
arXiv:2608.25717v1 Announce Type: new Abstract: Retrieval-augmented generation (RAG) is widely assumed to mitigate factual errors in large language models (LLMs), but it remains unclear whether retrieval uniformly compensates for missing knowledge. We study this question in a controlled factual QA setting over public companies, constructing a benchmark of approximately 2,000 firms across global equity indices. We evaluate six LLMs on four atomic attributes under four conditions: no-context, perfect context, misleading context, and distraction context. We find strong geographic disparities in no-context accuracy, indicating uneven parametric knowledge. While perfect context improves performance, it does not eliminate these gaps: gains are correlated with baseline accuracy, suggesting retrieval effectiveness is coupled to internal representations. Under misleading context, models frequently copy incorrect information. Larger models improve overall performance but do not remove these stru
arXiv:2608.25677v1 Announce Type: new Abstract: Parameter-efficient fine-tuning is often assumed to preserve pretrained capabilities because it updates only a small number of parameters. We show that this assumption depends strongly on adapter capacity. We study factual acquisition in a controlled OpenStreetMap-derived benchmark where Qwen3-4B must acquire anonymized geographic associations while retaining unrelated capabilities. Comparing full fine-tuning (FFT) with quantized low-rank adaptation (QLoRA) at ranks 8, 16, 32, and 64, we find that rank induces a clear acquisition--retention frontier. Low-rank QLoRA preserves out-of-domain (OOD) performance but acquires fewer facts, whereas higher ranks improve same-fact paraphrase generalization at an increasing cost in performance on unrelated benchmarks. FFT behaves as a conservative baseline: it retains general capabilities well, but does not reach the highest factual-acquisition regime. Distributional, weight-space, and spectral diagn
arXiv:2608.25662v1 Announce Type: new Abstract: In 2026, we held the fourth iteration of the SHROOM Shared Task series: SHROOM-Visions (\textbf{S}hared-task on \textbf{H}allucinations and \textbf{R}elated \textbf{O}bservable \textbf{O}vergeneration \textbf{M}istakes in \textbf{Vision} language model\textbf{s}), which is hosted at the UncertaiNLP Workshop co-located with EMNLP 2026. Following the success of the 2024 and 2025 tasks, this time we aim to tackle hallucinations through a model-agnostic detection task focused on large vision-language models. Building on the recently introduced SHEEP dataset, designed for long-term evaluation across model generations, the task invites participants to detect and classify fine-grained hallucination spans in image-conditioned text generation (VQA, image captioning, etc.). The evaluation uses a five-class taxonomy of hallucinations spanning four languages: Chinese, English, French, and Italian. The shared task generated strong interest in the NLP
arXiv:2608.25660v1 Announce Type: new Abstract: Automated novelty judgment can accelerate scientific discovery by enabling efficient evaluation, refinement, and comparison of research ideas. While large language models are increasingly adopted for this task, we investigate a previously overlooked limitation in their judgment capabilities: despite generating reasoning rationales that closely mirror those of human experts, their final novelty judgments often diverge substantially. We demonstrate that this miscalibration stems from a systematic bias towards judging ideas as "medium novel". To mitigate this, we propose Think-Probe-Respond (TPR), a lightweight approach that probes latent novelty judgments from hidden states during the reasoning phase and uses the probed judgments to condition the final response. Across strong baselines, TPR improves novelty judgment performance by 22.30% and successfully mitigates the prevalent "medium novelty" bias.
arXiv:2608.25655v1 Announce Type: new Abstract: Conversations with chat assistants increasingly span many topics in a single long-running thread, challenging memory systems. Existing long-context and memory benchmarks often expose session or topic boundaries, or probe direct personal-memory questions. These settings understate a harder assistant-memory regime: a flat mixed-topic thread where the system must infer which earlier episode makes a later task decision valid. We introduce SCALE-QA, a constraint-grounded task QA benchmark for flat unsegmented threads targeting episode integrity failure. The dataset contains 3,000 audited questions across 10 domains, uses deterministic four-way multiple-choice grading, and includes a deterministic runtime builder; experiments use all 3,000 questions through 128k and a stratified 400-question diagnostic at 1M. SCALE-QA questions are ordinary task-oriented requests whose correct answer depends on causally related evidence introduced earlier in th
arXiv:2608.25654v1 Announce Type: new Abstract: Open-ended Theory-of-Mind (ToM) trackers emit valid beliefs absent from finite references. A finite-reference-plus-matcher pipeline marks unmatched outputs false, creating proxy labels that can reverse proper-score model selection on fixed outputs. Holding 259 beliefs and paired scores fixed, reference recoding lowers weighted prevalence from 0.783 to 0.295 and reverses strictly proper Brier risk: a frozen source-prior rule leads native confidence by 0.227 under reference labels and trails by 0.152 under blinded adjudication, in all six authored scenarios. A reference-only Platt recalibrator reverses further. An ICE-specific reversal appears in a released 301-question NQ-open DPR-BERT pipeline: its average-confidence baseline improves instance-level calibration error by 0.045 under exact match but worsens it by 0.074 under human correctness, with both intervals excluding zero. On independently authored OpenToM narratives, 90-96% of audite
arXiv:2608.25637v1 Announce Type: new Abstract: Reference-based verifiers are important for evaluating reasoning models and providing accurate outcome rewards in reinforcement learning with verifiable rewards. To improve verification accuracy, prior work has explored rule-based, model-based, and tool-augmented verifiers for checking answer equivalence across diverse answer forms. However, the equivalence of answer forms such as $1+3.14$ and $1+\pi$ may depend on the question and scoring criterion. We frame such implicit assumptions as verifier inductive biases. To address this challenge, we propose AutoVerifier, a residual-guided non-parametric optimization method that learns these biases from recurring verifier errors. Specifically, AutoVerifier records these biases in rule cards and promotes them to code modules or prompt guidance only after replay validation detects no direct regressions, keeping accepted updates auditable, editable, and reusable. Experiments on four verifier benchm
arXiv:2608.25618v1 Announce Type: new Abstract: Long-document visual question answering increasingly relies on VLM agents that retrieve candidate pages, inspect page images, write findings to working memory, and synthesize answers. Working memory should carry answer-supporting evidence across page inspections for later grounded answering, yet existing evaluation mainly checks final-answer correctness and evidence-page access. This creates a memory-quality blind spot: an agent may reach the right page and answer correctly while leaving behind memory too generic or incomplete to support answering once page context is removed. We introduce \emph{memory-only answerability}, a diagnostic that asks whether a reader can answer from the question and terminal working memory alone. Building on this diagnostic, \emph{Answerable Working Memory} (AWM) treats terminal working memory as an answerable evidence artifact, and AWM-GRPO incorporates this signal into the GRPO reward while preserving final-
arXiv:2608.25605v1 Announce Type: new Abstract: Large language models (LLMs) demonstrate impressive performance across a wide range of general NLP tasks; however, their effectiveness in sensitive domains, such as hate speech detection, remains less clear. Prior studies comparing prompted LLMs with state-of-the-art encoder-based models (e.g., BERT variants (Roy et al., 2023; D\"onmez et al., 2024)) have shown only marginal gains, suggesting that LLMs may not excel in hate speech detection or mitigation. In this work, we revisit this question through the lens of instruction tuning. By thoroughly unifying 36 English hate speech datasets spanning multiple labeling schemes, we fine-tune a generalist LLM, based on Qwen3 (Qwen Team, 2025), specifically for hate speech mitigation. Our results demonstrate not only state-of-the-art performance on in-domain benchmarks but also substantial improvements in cross-domain and cross-lingual generalization--areas where encoder-based specialist classifie
arXiv:2608.25593v1 Announce Type: new Abstract: Agent capability is not determined by the model alone. The agent harness, encompassing memory management, planning strategy, action protocol, and tool/skill orchestration, can dominate the contribution of the underlying foundation model. Yet harness design remains manual, task-specific, and fundamentally unscalable. We present JIT-Agent, a harness intelligence model trained to synthesize task-adaptive agent harnesses on the fly for arbitrary off-the-shelf agentic LLMs. We formalize the agent harness as a composable, machine-generatable artifact governed by a fixed four-module protocol, and train JIT-Agent to customize harnesses for a given task at hand, repair harnesses for stable and reliable execution, and self-evolve by distilling performance signals from an expanding archive of prior harness configurations. Equipped with JIT-Agent as a harness helper, DeepSeek-V4-Flash surpasses GPT-5.6 on DeepSearchQA (+9.1) and OdysseyBench (+4.3),
arXiv:2608.25583v1 Announce Type: new Abstract: Reasoning-oriented large language models often achieve strong problem-solving performance by generating long chains of thought, but this behavior substantially increases inference cost and latency. In contrast, instruction-tuned models tend to answer more concisely, yet often lack comparable reasoning ability. This accuracy-efficiency mismatch motivates a lightweight approach that combines the strengths of both models without full model retraining. In this paper, we propose GRIP (Granular Reward-guided Interpolation of Parameters), a reward-guided parameter interpolation framework for efficient reasoning. Given a reasoning model and an instruction model with identical architectures, GRIP assigns learnable interpolation ratios to individual modules and optimizes only these ratios while keeping both source models frozen. The interpolation ratios are trained with a reward signal that favors responses that are both correct and concise. Experi
arXiv:2608.25579v1 Announce Type: new Abstract: Metaphor-identification performance can change markedly across datasets that differ in text distribution and annotation policy. We examine whether a fixed expert-informed procedure produces a more even cross-dataset profile than task-specific parameter adaptation. Four prespecified conditions are compared for Chinese sentence-level metaphor identification: BERT fine-tuning (BERT-FT), QLoRA-based large language model fine-tuning (LLM-FT), direct zero-shot LLM prompting (LLM-ZS), and zero-shot prompting with a frozen procedural Skill (Skill-ZS). The Skill operationalizes established criteria involving contextual meaning, basic meaning, contrast, and comparison. Evaluation covers CMRE Test and two external datasets, CCIME and CMC. Fine-tuned scores are means over three seeds, whereas each zero-shot score comes from one deterministic configuration. Fine-tuning remains strongest on the native test set: BERT-FT reaches 91.76 Macro-F1. LLM-FT ha
arXiv:2608.25574v1 Announce Type: new Abstract: Automatic Speech Recognition (ASR) is typically evaluated using Word Error Rate (WER), which poorly reflects semantic similarity. While embedding-based metrics correlate better with human judgments, the respective roles of encoder and decoder-based Large Language Models (LLMs) remain underexplored. This paper presents a comparative study of both families for ASR evaluation. We analyze BERTScore and SemDist across different LLMs, layers, and pooling strategies, showing that both metrics can achieve strong correlation with human judgments when properly configured. For decoder models, we investigate generative LLMs in two settings: pairwise hypothesis selection via prompting and direct qualitative error classification. Our results show that encoder-based metrics remain highly competitive, while generative LLMs perform strongly in hypothesis comparison and improve the interpretability of ASR evaluation.
arXiv:2608.25569v1 Announce Type: new Abstract: Large Language Models (LLMs) often produce emotionally flattened responses after alignment, limiting their effectiveness in affect-sensitive applications. In this paper, we propose EmoVec, a lightweight framework for controllable affective generation via latent vector steering. EmoVec extracts emotion-specific directions from paired neutral and emotion-conditioned responses using contrastive activation addition, and further refines them through task-specific debiasing and principal subspace removal. During inference, these vectors are injected into the final residual stream with static or scenario-adaptive scaling, enabling continuous control over emotional intensity without updating model weights. Experiments across three LLMs and eight emotions show that EmoVec consistently improves emotional salience while largely preserving semantic content, fluency, and coherence. Ablation studies and human evaluation further confirm the effectivenes
arXiv:2608.25561v1 Announce Type: new Abstract: VLMs are increasingly positioned as daily assistants that perceive first-person environments, follow user dialogue, and decide how to help. Existing egocentric benchmarks mainly evaluate visual understanding in isolation, leaving open whether models can arbitrate between visual evidence and user-provided language when the two are helpful, irrelevant, or conflicting. We introduce EgoArgus, a human-annotated dataset for evaluating egocentric assistants on understanding and decision tasks in five dialogue-video daily scenarios. Our results demonstrate that it is still challenging for current VLMs as reliable egocentric assistants, which requires identifying which modality is trustworthy and deciding when intervention is warranted. Deeper analysis also shows that existing modality bias mitigation methods are quite restricted to enhance performance, providing insights to aid practioners into the deployment of current VLMs as daily assistants.
arXiv:2608.25555v1 Announce Type: new Abstract: Explainability for transformer-based language models is becoming crucial as these systems are deployed in high-stakes applications. As a result, the ecosystem of explainability tools is rapidly evolving, becoming richer, but also more fragmented and harder to navigate. To address this challenge, we present Virgil, an interactive system that lets practitioners and researchers, including non-experts, navigate explainability tools for transformer language models. Supported by a curated knowledge base, the system enables users to discover and compare explainability tools within a unified interface.
arXiv:2608.25531v1 Announce Type: new Abstract: Humanities and social science research requires close reading of long narrative materials such as novels, scripts, archives, and case reports, yet many users have limited access to costly proprietary long-context models. Compact, locally deployable language models are a practical alternative, but directly feeding them an entire long context remains costly, hard to inspect, and prone to missing sparse evidence. We present ClueWeaver, an evidence-aware dual-agent framework for long-narrative question answering with compact local models. A Finder identifies passages containing answer-critical clues through retrieval-guided segmentation, while an Interpreter derives the answer from the selected evidence, produces rationales with paragraph-ID citations, and applies an internal self-calibration pass for high-risk questions. Both agents are optimized with reward-guided reinforcement learning: Finder rewards emphasize evidence retention and faith
arXiv:2608.25523v1 Announce Type: new Abstract: Prefix caching introduces a fundamental tradeoff in multi-agent large language model (LLM) serving: retaining a long system-prompt key-value (KV) cache for an agent accelerates future calls, yet it reduces the GPU memory available for batching concurrent requests. In multi-stage workflows, existing schedulers tend to prioritize either immediate prefix locality or overall workflow progress. However, under a shared KV cache budget, optimizing either objective in isolation can prolong tasklevel job completion time (JCT) through downstream delays or frequent prefix replacement. To strike a balance, we here propose TOPAS, a Task-Oriented Prefix-Aware Scheduler that jointly decides which agent prefixes to keep in the cache and which requests to schedule for execution. TOPAS scores candidate post-decision states by trading off the expected reduction in each task's longest remaining service path against the near-term benefit of downstream prefix
arXiv:2608.25487v1 Announce Type: new Abstract: Retrieval-Augmented Generation (RAG) has emerged as a powerful architecture for Question Answering (QA) by integrating external information into Large Language Models (LLMs). However, false, inaccurate, and misleading information in news and social media poses a serious challenge to real-world RAG systems, especially in multi-hop QA, where complex multi-step reasoning can be misled by even a single deceptive misinformation segment in the retrieved documents. Existing approaches mainly rely on implicit alignment or explicit regulation, but their limited ability to assess fine-grained information reliability makes them vulnerable to deceptive misinformation that is semantically relevant to the question yet factually incorrect, leading to erroneous answers. To address this limitation, we propose ReliableRAG, which, to the best of our knowledge, is the first reliability-driven framework that mitigates deceptive misinformation in multi-hop QA
arXiv:2608.25478v1 Announce Type: new Abstract: In recent years, distinguishing between AI-generated text and human-written text has remained a challenge. In this paper, we introduce VietAIDetector, an open-source tool designed specifically for detecting Vietnamese AI-generated text. It allows users to interact through a Gradio web interface with inputs ranging from raw Vietnamese text to common text file formats, including scanned documents and exceptionally long texts that exceed the context size of the employed Large Language Models (LLMs). The core component of the tool employs a Zero-Shot approach to detect AI-generated text without requiring domain-specific training data, building upon the previous VietBinoculars and Binoculars research. The tool is built upon a Vietnamese-specific language model and has been evaluated on out-of-domain datasets, demonstrating superior performance compared to existing methods primarily developed for English. Additionally, users can select optimal
arXiv:2608.25449v1 Announce Type: new Abstract: Formal theorem proving enables machine-verifiable evaluation of mathematical reasoning, yet existing benchmarks often emphasize aggregate proof accuracy, concentrate on a narrow range of mathematics, and provide limited evidence of robustness to equivalent reformulations. We introduce MathAdv, a diagnostic benchmark spanning 13 domains across undergraduate- and graduate-level mathematics. Alongside Lean 4 theorem proving, MathAdv provides up to three auxiliary tasks: multiple-choice questions that probe mathematical knowledge, fill-in-the-blank problems that isolate informal reasoning, and expert-crafted transformations that test robustness to problem presentation. Our evaluation of contemporary theorem provers yields four findings: formalization remains a major bottleneck; performance varies substantially across mathematical domains; natural-language guidance helps general-purpose LLMs but can hinder proof-specialized models; and mathema
arXiv:2608.25428v1 Announce Type: new Abstract: Correcting flawed reasoning traces remains a significant challenge for Large Language Models (LLMs), whose autoregressive generation can propagate early mistakes into subsequent reasoning. We introduce DCGC, a Masked Diffusion Model (MDM) framework for global correction that uses an imperfect solution draft from an upstream solver as auxiliary context. DCGC combines task-specific Supervised Fine-Tuning (SFT) with a novel inference-time mechanism called Dynamic Dual-CFG. This mechanism separates problem-only and joint problem-draft branches and scales the draft-conditioned residual using a relative confidence gap. Across math, code, and knowledge reasoning benchmarks, DCGC outperforms standard sampling and simpler CFG variants, with additional results suggesting transfer to different diffusion backbones. In test-time setting where ground-truth failure labels are unavailable, DCGC improves full test set accuracy by correcting low-consensus
arXiv:2608.25398v1 Announce Type: new Abstract: Multimodal Large Language Models (MLLMs) have demonstrated strong abilities in solving diverse visual and textual reasoning tasks. However, their development in the physics domain is significantly hindered by the lack of a comprehensive benchmark. To fill this gap, we introduce OmniPhys, a large-scale benchmark for multimodal physics understanding and reasoning, covering middle school through university-level problems from Chinese Educational Corpora. OmniPhys consists of 15,246 questions and 19,850 images, accompanied by detailed annotations that support fine-grained analysis of reasoning processes and knowledge usage. Beyond conventional evaluation, OmniPhys is a benchmark that systematically evaluates multimodal outputs in the physics domain, including models' ability to generate structured physics diagrams, which constitute a fundamental component of authentic physics problem solving. Extensive evaluations reveal critical gaps in the
arXiv:2608.25379v1 Announce Type: new Abstract: Chain-of-thought prompting can expose and amplify demographic stereotypes within an LLM's intermediate reasoning and create a failure mode that final-answer debiasing alone cannot address. Mitigating such bias during generation presents a fundamental timing problem: intervening too late allows biased reasoning to propagate, while unnecessarily intervening can disrupt otherwise correct reasoning. Existing approaches largely avoid this decision by either evaluating completed reasoning chains post hoc or intervening at predetermined steps, leaving open when a developing reasoning trajectory provides sufficient evidence to warrant correction. We formulate this decision as an online change-point detection problem. A per-step bias signal updates a CUSUM statistic and a targeted correction is injected only when accumulated evidence crosses a detector-specific threshold calibrated on held-out data. We instantiate the framework with a white-box si
arXiv:2608.25359v1 Announce Type: new Abstract: Conversational derailment forecasting aims to predict when online discussions will escalate into hostility, enabling proactive moderation. Existing approaches often struggle in low-data settings and to generalize across domains. This poses a challenge for new platforms and smaller communities where annotated data is limited. We propose modeling pragmatic representations of conversations to reduce lexical noise and improve generalizability. Specifically, speech act information is used as an auxiliary learning signal alongside textual semantics. Experimental results show improved performance across three datasets, particularly in low-data and cross-domain settings.
arXiv:2608.25347v1 Announce Type: new Abstract: The Jacobian lens (J-lens) has been proposed as a way to read verbalizable representations from language models. However, its principle and meaning lack a detailed and theoretical discussion. We provide a mathematical view of this interpretation and of its assumed causal structure. Besides treating the J-lens as a heuristic probe, we further regard it as a first-order causal transfer operator from intermediate activations to expected future readouts. We study the Jacobian matrix as the optimal local linear approximation of the downstream mapping, analyze its global approximation behavior and bias, and identify its mathematical meaning as an expectation over anticipated future readouts. Further analysis of the Jacobian energy distribution reveals that its causal geometry is highly sparse. The energy decays with depth, concentrates in an extremely small proportion, and decomposes into diagonal pathways and specific critical positions. This
arXiv:2608.25343v1 Announce Type: new Abstract: Chinese query correction (CQC) is important for search and query recommendation on content platforms, but supervised methods rely on large annotated correction pairs that are costly to maintain as query vocabularies evolve. Unsupervised correction with language models is attractive, yet in the short-query setting, unconstrained generation often over-corrects ambiguous inputs toward high-frequency phrases, causing intent drift. We propose \textsc{GUIDE}, a generative unsupervised framework for CQC based on a confuse-then-clarify paradigm. \textsc{GUIDE} encodes phonetically or visually confusable characters with shared-IDs and reconstructs the original query with an encoder--decoder architecture, which constrains correction to plausible confusion neighborhoods while learning from unlabeled query streams. A time-decayed, query-frequency-weighted objective further supports adaptation to rapidly changing query vocabularies. Experiments on \te
arXiv:2608.25336v1 Announce Type: new Abstract: Large language models (LLMs) can fluently verbalize statistical evidence, yet statistical reports can still drift numerical values, invert effect directions, or restate thresholded contrasts as categorical effects. We frame these failures as a control problem: the evidence-bearing content of a scientific report should be fixed by structured statistical results rather than sampled during prose generation. We therefore use cross-run reproducibility to stress-test whether report-visible numbers and claims are bound before prose generation. Existing controls operate at the text or slot level; a deterministic hybrid template reproduces only 61.1% of report-visible numerical content across seeds because the LLM still selects which findings and numbers the template renders. We propose claim-locked reporting, a provenance-before-prose protocol that fixes the evidence source, numbers, direction, and allowed language strength of each reportable cla
arXiv:2608.25277v1 Announce Type: new Abstract: Multi-agent LLM systems coordinate through natural-language messages that consume 40--60\% of their token budget. Replacing these with structured graphs reduces cost but fails on tasks requiring adaptive reasoning. We propose \textbf{Routed Graph Handoff}, where a lightweight LLM router (155 tokens, 0.15\% overhead) selects between a typed dependency graph and natural language for each delegation. On four benchmarks (1,050+ trajectories), the routed system matches or exceeds NL-only on every task: \textbf{+12.7\,pp} on $\tau$-retail at 3.2$\times$ compression ($p{<}0.01$), \textbf{+8.7\,pp} on BrowseComp at 2.2$\times$ compression ($p{<}0.05$), and parity on BFCL and AppWorld. Without the router, graph-only delegation regresses 14.6\,pp on AppWorld; the router eliminates this at near-zero cost. A graph-aware executor prompt is required: the same schema without interpretation guidance yields no gain. An oracle analysis reveals 8.6\,pp of a
arXiv:2608.25276v1 Announce Type: new Abstract: Mixture-of-Experts (MoE) architectures enable scalable and efficient large language models (LLMs) by selectively activating expert sub-networks through a routing mechanism. However, this adaptive design introduces a new attack surface: specific experts become disproportionately correlated with certain tokens (e.g., end-of-sequence), allowing adversaries to manipulate model behavior via lightweight perturbations. In this work, we present \textbf{Groundhog Bit-Flip Attack (GBFA)}, the first bit-flip-based \textit{ Denial-of-Wallet availability attack} against MoE-based LLMs. By identifying and flipping routing-layer bits associated with related expert activations, we demonstrate that GBFA substantially extends the decoding token usage across three different LLM modes: conversational, reasoning, and agentic tasks, while largely preserving semantic fidelity. Across four main real-world MoE-based LLMs, manually deactivating on average fewer th
arXiv:2608.25243v1 Announce Type: new Abstract: Continual knowledge injection is essential for keeping large language models up-to-date in a fast-evolving world. Existing methods rely on supervised fine-tuning (SFT), which memorizes injected facts in their training format but fails to generalize across paraphrasing, document combinations, and reasoning. To address this, we propose Golden-GRPO Injection (GRIN), a three-stage self-learning framework for continual knowledge injection. Golden-GRPO is a mixed-policy reinforcement learning algorithm designed specifically for knowledge injection, which injects a golden answer to provide learning signal even when on-policy rollouts fail on novel facts. We further introduce Blank and Counter, two document-level benchmarks targeting novel acquisition and counterfactual overwrite respectively, each evaluating single-fact recall, multi-source retrieval, and inferential reasoning. Our experiments establish a clear empirical claim: mixed-policy rein
arXiv:2608.25190v1 Announce Type: new Abstract: Fake news detection has become an important Natural Language Processing (NLP) task due to the rapid spread of misinformation through online news platforms and social media. While transformer-based models such as BanglaBERT achieve strong performance for Bangla text classification, their quadratic computational complexity makes them less suitable for long-document processing in resource-constrained environments. This paper investigates Mamba-based State Space Models (SSMs) as an efficient alternative for Bangla fake news detection. We propose BanglaMamba and compare it with pre-trained BanglaBERT and a similarly configured BERT model trained from scratch. Experimental results show that BanglaBERT achieves the highest Macro-F1 score (0.9260), while BanglaMamba (0.9029) achieves performance comparable to the from-scratch CustomBERT (0.9057) despite using a different architecture. Meanwhile, BanglaMamba achieves approximately $2.2\times$ high
arXiv:2608.25166v1 Announce Type: new Abstract: Transformer representations describe trajectories through high-dimensional vector spaces, which are shaped dynamically as tokens incorporate relational context across layers. Such data tend to concentrate on lower-dimensional sub-manifolds, a form of compression quantified by the Intrinsic Dimensionality (ID), the minimum number of independent variables needed to represent them without significant information loss. In this work, we ask whether the grammatical role of tokens, as marked by their part-of-speech (PoS) tag, shapes the local geometry of this manifold. To this end: (1) We investigate the layer-wise evolution of ID, finding that closed-class items expand earlier and collapse sooner than open-class ones; (2) We show its expansion and contraction to be explained by changes in the neighborhood structure, and hence in the relations between words within a sentence; (3) We compare encoders (ModernBERT, bigbird-roberta-large) and decode
arXiv:2608.25152v1 Announce Type: new Abstract: Multi-agent LLM systems increasingly debate answers, coordinate research, simulate users, and mediate information flows, making agent-to-agent persuasion a basic but undermeasured capability. We introduce a controlled testbed for studying how goal-directed persuaders shift elicited stances in networks of LLM agents grounded in real-world ego-network topologies. Across four LLM backbones, five graphs, and 55 policy statements, we find that persuasion dynamics depend on the interaction between topology, competition, topic, and model prior. Additionally, we show that direct exposure reliably predicts next-round stance change in competing runs, and peer relays carry smaller but measurable influence, showing that agents not assigned to persuade can still transmit persuasive force. Finally, analyzing post text alone misses important movement: planned strategies are only partly realized in executed messages, action choices can diverge from messa
arXiv:2608.25123v1 Announce Type: new Abstract: Retrieval-augmented generation (RAG) improves large language models by incorporating external knowledge without retraining, but existing methods often underuse the relational structure encoded in knowledge graphs. Graph-based RAG can capture entity relationships, yet supervised graph retrieval typically requires labeled question-answer data that may not be available for newly constructed graphs. We address this limitation with SelfGraphRAG, a framework that generates question-answer pairs directly from knowledge graph structure and uses them to train a query-conditioned graph retriever. The generated questions capture multi-hop paths and local neighborhoods, providing relational supervision without manual annotation. Experiments on multi-hop question answering and classification benchmarks show that SelfGraphRAG improves retrieval precision and downstream reasoning performance over embedding-based baselines. These results suggest that kno
arXiv:2608.25115v1 Announce Type: new Abstract: Existing methods for improving Retrieval-Augmented Generation (RAG) efficiency mainly optimize downstream LLM generation, such as context compression or serving optimization. However, RAG is an end-to-end system, and its bottleneck can shift between upstream reranking and downstream generation under different serving loads and reranking budgets.In this paper, we first empirically characterize this shifting-bottleneck behavior and show that upstream reranking can become the dominant bottleneck under high query rates or large reranking budgets. Reducing the reranking budget can relieve this bottleneck, but it may also drop supporting evidence and degrade recall. To address this problem, we propose \textbf{\textsf{PACE}} (\textbf{P}rioritized \textbf{A}daptive \textbf{C}overage of \textbf{E}vidence), a training-free framework that combines \textit{evidence frontloading} with \textit{pressure-adaptive budgeting}. \textsf{PACE} first reorders
arXiv:2608.25089v1 Announce Type: new Abstract: Crosslingual evaluation of language models that enables fair comparisons remains a fundamental challenge in multilingual NLP. Existing studies adopt a variety of downstream tasks and intrinsic metrics with different theoretical justifications, yet there has been little empirical investigation into whether these approaches yield meaningful crosslingual conclusions. We systematically examine crosslingual evaluation approaches using controlled monolingual language models trained on parallel data with varying tokenizer vocabulary sizes and model sizes, and further validate our findings on multilingual LLMs. We further discuss challenges in achieving comparable downstream evaluation across languages. Our results show that several widely used normalized metrics introduce crosslinguistic biases rooted in tokenization, encoding, and orthographic differences. In contrast, sentence-level negative log-likelihood computed over semantically equivalent
arXiv:2608.25085v1 Announce Type: new Abstract: Clinical diagnosis is fundamentally interactive and incremental, yet the dominant paradigm for evaluating Large Language Models (LLMs) in medicine remains static QA benchmarks or template-based dialogues. These benchmarks say little about whether a model can serve as a diagnostic agent in a dynamic clinical encounter, with LLMs showing significant accuracy and reliability degradation in multi-turn settings. To address this issue, we present MTDiag, a large multi-turn diagnostic dialogue dataset constructed from three heterogeneous sources: DDXPlus, MIMIC-IV, and published case reports (AJCR), covering common ED presentations as well as long-tail rare and atypical conditions. All cases are normalized into a canonical schema anchored in the most comprehensive and widely-adopted medical knowledge bases (UMLS concept identifiers, with ICD-10 diagnosis codes). We release the schema, a UserLM-8B-based utterance-generation pipeline, and the phys
arXiv:2608.25061v1 Announce Type: new Abstract: GPUs increasingly accelerate database systems, but query-specific peak performance still often relies on hand-written kernels. Existing LLM kernel benchmarks focus on machine learning operators, leaving irregular, heterogeneous, data-movement-heavy database-style operators untested. We introduce DataKernelBench, which translates SQL into validated PyTorch TorchPlan programs and evaluates LLMs that optimize either the core tensor-bounded snippet or the full query in CUDA or Triton through execution-guided repair. Across ten proprietary and open-weight models on TPC-H SF10 with an H100 GPU, the strongest full-query CUDA configuration achieves $2.11\times$ speedup over torch.compile at full pass rate. We find that higher-performing implementations commonly use kernel fusion and execution-strategy changes, stronger models benefit most from full-query specialization, and workload context matters more than hardware context. To handle data large
arXiv:2608.25038v1 Announce Type: new Abstract: We introduce grounded glossary generation, a structured task requiring models to recover semantically meaningful Sanskrit phrases and produce translation-grounded meanings from a sloka-translation pair, formalizing the traditional patha commentary practice as an evaluable NLP objective. We construct a benchmark of 31,316 sloka-translation-glossary triples from the Valmiki Ramayana and Srimad Bhagavatam, paired with two metrics: Jaccard for phrase recovery and Meaning Faithfulness for semantic consistency. Across zero-shot, few-shot, and instruction fine-tuned variants of Gemma-3n-E4B, Gemma-3-12B, Phi-4, and Qwen3.5-9B, instruction fine-tuning substantially outperforms prompting, while explicit segmentation yields gains. Error analysis identifies over-segmentation of sandhi and samasa compounds as the dominant failure mode, pointing to morphological modeling as the key bottleneck for faithful Sanskrit lexical decomposition.
arXiv:2608.25028v1 Announce Type: new Abstract: Spoken-language analysis via prompt-based domain-adaptive models is a promising direction for low-resource, non-invasive dementia screening, but such models remain internally opaque. We study the interpretability of the Domain-Adapted models via Prompt-based Fine-tuning (DAPF) framework, which casts dementia detection as diagnosis-related masked-token prediction. We interpret DAPF and strong baselines using a variety of probing and analysis techniques, finding that DAPF achieved the best overall performance (accuracy=0.83 and macro-F1=0.83) with diagnosis most recoverable from its [MASK] representation. However, this representational advantage did not extend to token-level explanation faithfulness. DAPF attributions primarily reflected language task vocabulary, discourse markers, and transcription artifacts, with perturbation tests showing weak or negative effects. This suggests that its masked-token interface determines diagnosis informa
arXiv:2608.25022v1 Announce Type: new Abstract: As people adopt transformer-based language models (e.g., ChatGPT and Gemini) for an increasing number of use-cases, it is important to know how such models learn and represent the meaning of the language, and to be more informed about what language is. This document is an attempt to help the reader understand how linguistic meaning (i.e., semantics) is approached from different fields of scientific and philosophical examination. I also explain three primary semantic theories: formal semantics, grounded semantics, and distributional semantics then compare how transformer-based language models differ from how humans learn language.
arXiv:2608.25005v1 Announce Type: new Abstract: The imperfective paradox provides a useful test of compositional semantic analysis. Recent work constructs an NLI benchmark and reports that models frequently infer completed telic events from progressive descriptions, attributing this behavior to a Teleological Bias. It further argues that prompting interventions cause a Calibration Crisis. We reexamine the benchmark and conclusions and show that it is substantially affected by conceptual and evaluation mis-specifications. We identify three conceptual mis-specifications. In particular, Aspectual Reduction affects the benchmark construction, analysis, experiments, and conclusions. Under a strict NLI standard, 76% of Group A instances do not explicitly rule out culmination. In our native-speaker annotation, 38% of Group A examples and 29% of the Group C examples were judged to permit an alternative interpretation. To control these issues and lexical variation, we construct Lexically Matche
arXiv:2608.24988v1 Announce Type: new Abstract: Activation steering can be embedded directly into a language model's weights, shaping behaviour without inference-time intervention and offering a way to encode alignment prior to release. However, models are routinely fine-tuned after deployment, and it is unknown whether embedded interventions survive this. We study the stability of embedded steering for refusal suppression and brevity induction across five instruction-tuned models (3B-14B) under non-adversarial SFT and RLHF. Behaviourally, preservation tracks the training data: steering degrades when optimisation pressure contradicts the targeted behaviour and persists otherwise, with refusal ablation losing 64% of its effect on average under SFT. Mechanistically, however, the weight edit survives almost untouched even where behaviour reverts: mean vector recovery is $\rho = 0.004$, and the fine-tuning update along the steering direction is near-orthogonal to its pre-edit weight patter
arXiv:2608.24982v1 Announce Type: new Abstract: Foundation-model post-training usually relies on human labels, preference data, stronger teachers, or executable verifiers. We study Unsupervised Post-Training (UPT): update-bearing adaptation on unlabeled inputs whose learning signal is derived from same-lineage model artifacts rather than an external oracle. We catalog 80 strict UPT methods and organize them by the object that supplies the update signal: a prediction statistic, a sample relation, a self-generated target, or an internal evaluator. Beyond inventory, we show how the choice of internal signal and task structure determines whether post-training improves the model or recursively amplifies error. An orthogonal Input Visibility $\times$ Update Persistence view maps deployment regimes and defines a unified framework for UPT selection and evaluation.
arXiv:2608.24952v1 Announce Type: new Abstract: Systematic dialectal performance gaps in language models (LMs) are well documented, but the source of these disparities within the modern language modeling pipeline remains unclear. Our study traces this "dialect tax" across the natural language processing pipeline. Using parallel English dialect corpora that hold meaning fixed while varying surface form, we first confirm that LMs recognize matched Standard American English (SAE) and dialectal texts as semantically equivalent. However, we discover further representational gaps corresponding to downstream performance gaps. Across model families and generations, modern LMs still encode dialectal texts unequally during tokenization, pre-training, post-training, and inference. Strikingly, bypassing traditional subword segmentation via a character-level counterfactual tokenizer removes neither input and output asymmetries nor dialectal accuracy gaps. During pre-training, dialect pairs induce m
arXiv:2608.23137v2 Announce Type: replace-cross Abstract: Existing articulatory corpora based on real-time MRI and electromagnetic articulography capture tongue shape and motion but do not provide traceable labels for the muscle-driven process that generated an observed configuration. We introduce a simulator-grounded data-construction framework and instantiate it as 3DTongueQA. Controlled 11-dimensional muscle activations are mapped to fixed-topology tongue meshes with the ArtiSynth Badin finite-element model, converted into structured biomechanical records, and rendered as deterministic QA on muscle state, geometry, and target-directed change. We screen 295,157 configurations, retain 295,115 valid meshes, and construct 891,156 QA records per language. Language naturalization changes only surface form and is verified against the source records; English and Korean instantiations demonstrate construction-level portability. A swappable SpiralNet++--Qwen3-8B baseline reaches 62.9 $\pm$ 9.