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:2606.15974v2 Announce Type: replace Abstract: Despite the significant advancement of LLMs in conversation summarization, their evaluation remains limited by insufficient scenarios, input lengths, and sample sizes. Furthermore, existing benchmarks often omit frontier reasoning systems and efficient small models, or lack fine-grained, multi-dimensional assessments. To bridge these gaps, we propose OmniCSEval, a unified benchmark comprising 1,800 diverse conversations across six real-world scenarios, featuring context lengths ranging from 128 to 32k tokens. For fine-grained evaluation, we employ a bidirectional fact-checking framework that integrates key fact matching to assess completeness and conciseness, alongside summary fact verification to evaluate faithfulness. To ensure reliable assessment, we establish a human-LLM collaborative pipeline for key fact extraction and a multi-LLM consensus verifier for summary fact decomposition. Leveraging this framework, we evaluate 28 LLMs a
arXiv:2606.15412v2 Announce Type: replace Abstract: Biomedical relation extraction (BioRE) is a key step in transforming biomedical literature into structured knowledge. Most existing approaches rely on supervised models trained on costly annotated datasets, limiting their scalability and adaptability across relation types and domains. We investigate few-shot BioRE using prompt-based learning with large language models (LLMs) and compare two task formulations: pairwise classification, which predicts relations for individual entity pairs, and joint generation, which extracts multiple relations in a single model call. Experiments on the BioREDirect dataset reveal a clear precision-recall trade-off. Pairwise classification achieves higher recall, whereas joint generation is more precise and computationally efficient. The best-performing model achieves a micro-F1 score of 0.44, substantially outperforming previous few-shot results (0.34) while remaining below the supervised baseline (0.56)
arXiv:2606.13993v2 Announce Type: replace Abstract: A crucial aspect of linguistic capability is the ability to trade off between stored representations and abstract knowledge: one must retrieve learned representations, but also generate novel ones by applying productive rules. While recent work has examined abstract knowledge in language models, holistic storage of multi-word units has received far less attention. We probe internal representations in text-based LLMs and an ASR model, testing whether V+up phrasal verbs develop distinct representations as a function of frequency and predictability. All models show evidence of holistic storage driven by frequency and predictability, further supporting usage-based theories of language.
arXiv:2606.08417v2 Announce Type: replace Abstract: Diffusion and continuous flow-based language models have emerged as the leading non-autoregressive alternatives to language modeling. Progress in both paradigms is overwhelmingly tracked by generative perplexity (gen-PPL): the per-token negative log-likelihood of samples under a frozen autoregressive (AR) scorer such as gpt2-large, typically paired with an empirical-entropy guardrail to rule out low-entropy collapse. We argue that this metric is unsound. By construction, gen-PPL measures only predictability under the scoring AR, not grammaticality or semantic coherence -- and the set of predictable but still low-quality sequences is combinatorially large. To make this concrete, we construct a suite of zero-parameter, deliberately naive samplers that achieve state-of-the-art gen-PPL on LM1B and OpenWebText at non-degenerate entropy, surpassing recently published diffusion and continuous-flow models while producing text that is incohere
arXiv:2606.08011v3 Announce Type: replace Abstract: Prior work has explored prompting large language models (LLMs) to rewrite source text before translation, with the goal of improving machine translation (MT) quality. However, we find that such prompt-based rewriting can degrade translation quality rather than enhance it, particularly when smaller LLMs, such as 4B-parameter models, are used. We argue that this limitation stems from the difficulty of controlling rewriting behavior through natural-language prompts alone: a rewrite is useful only if it leads to a better downstream translation, yet existing prompt-based methods do not explicitly optimize for this signal. To address this issue, we propose \textbf{RLSR} (\textbf{R}einforcement \textbf{L}earning for \textbf{S}ource \textbf{R}ewriting), a reinforcement learning framework that trains the rewriting model with a reward derived from the downstream translation-quality improvement produced by each rewrite. Experiments across six MT
arXiv:2606.05158v2 Announce Type: replace Abstract: Multi-agent reasoning systems adopt a "generate-then-transfer" paradigm that forces end-to-end latency to scale linearly with pipeline depth. We introduce StreamMA, a multi-agent reasoning system that streams each reasoning step to downstream agents as soon as it is generated, pipelining adjacent agents and thus reducing latency. Surprisingly, this pipelining also improves effectiveness: because multi-step reasoning quality is non-uniform and early steps are more reliable than later ones, working with these reliable early steps instead of the full chain prevents error-prone late steps from misleading downstream agents. We formalize both advantages with the first closed-form joint analysis of stream, serial, and single protocols, deriving the effectiveness ordering, speedup upper bound, and cost ratio. Across eight reasoning benchmarks spanning mathematics, science, and code, two frontier LLMs (Claude Opus 4.6 and GPT-5.4), and three t
arXiv:2605.26352v2 Announce Type: replace Abstract: Retrieval is increasingly moving from one-shot matching toward interactive reasoning, where language agents iteratively inspect evidence, reformulate queries, and search again. Training such agents raises a credit-assignment challenge: executable actions such as queries or summaries can be directly evaluated by the retriever, while latent reasoning steps are not directly observable and only affect future executable actions. This asymmetry makes outcome-level reward assignment unreliable, as the same final reward may credit reasoning steps that did not actually shape retrieval success. We propose RICE-PO, a critic-free policy optimization framework that converts retrieval interactions into localized learning signals. RICE-PO selects high-uncertainty executable actions as anchors, evaluates local counterfactual branches using retrieval metrics, and propagates credit to latent reasoning steps only when reasoning-to-action influence is st
arXiv:2605.26045v2 Announce Type: replace Abstract: An activation oracle is a language model trained to read another model's internal activations and describe them in natural language, for example to name a secret word the other model was trained to hide. Oracle answers carry no measure of confidence, which limits their use in auditing. We compare five ways of attaching a confidence score to an oracle's answer on this secret-word task, across four oracles from two model families (Qwen and Gemma, 8B to 27B parameters), at $6{,}000$ samples per method and oracle. The five methods rank the same way on all four oracles. Which method to use depends on one question: can the auditor list the possible answers in advance? If the auditor can, then having the oracle score each candidate answer roughly doubles accuracy and separates correct from wrong answers best of the five (AUROC $0.92$ to $0.96$). If the oracle must generate its answer freely and no labeled data exists, the agreement rate over
arXiv:2605.25596v2 Announce Type: replace Abstract: Phonological features provide a language-general and linguistically grounded representation of speech. We present PhonoQ-2.0, a multilingual frame-level phonological feature recognizer built on self-supervised speech models. The system directly predicts a structured 22-dimensional feature vector per frame encoding manner, vowel quality, place, and voicing, instead of deriving features from phoneme outputs. To ensure phonologically coherent predictions, we introduce a manner-conditioned gating mechanism that activates valid feature groups. Evaluated across multiple languages and corpora, PhonoQ-2.0 achieves an average macro-F1 of 91.3% in-domain and 88.9% out-of-domain. Compared to a strong CTC phoneme baseline, it delivers consistent gains of +8.8 F1 in-domain and +8.6 out-of-domain on average. In unseen-language evaluation, PhonoQ-2.0 improves macro-F1 from 66.9% to 73.6% (+6.7 on average), with gains of up to +10.8 points.
arXiv:2605.25344v2 Announce Type: replace Abstract: Dense linear maps carry much of the parameter and computational burden of modern neural networks, yet their dense form leaves the organization of learned couplings implicit. Quantum many-body physics organizes exponentially large operators by writing a global Hamiltonian as a sum of local terms, \(\hat H=\sum_k\hat h_k\). Whether the same structural principle can carry learned neural maps is unknown. We introduce Tensor Mixture (MixT), which represents a dense map as a natively executable sum of overlapping local tensor operators without imposing an explicit matrix-rank constraint. The local-term count \(N_T\) sets the effective nonlocality and operator complexity, while the number of replaced Transformer blocks \(N_B\) extends this structural coordinate across network depth. Tests on Qwen3-8B and LLaMA2-7B reveal a broad recoverable regime followed by an abrupt, model-specific boundary that is remarkably stable against changes in \(N
arXiv:2605.23668v2 Announce Type: replace Abstract: Although large language model (LLM) conversational systems process millions of multi-turn dialogues daily, they remain fundamentally reactive: they respond only after the user types a query. A key step toward proactive interaction is next-query prediction, which anticipates the user's subsequent query based solely on the preceding dialogue. Progress on this task is hindered by the lack of dedicated benchmarks and a fundamental efficiency--quality trade-off: naively concatenating full dialogue history incurs linearly growing token consumption, while truncating to the latest turn discards crucial cross-turn context. Our key insight is that accurate prediction does not require re-reading raw history; it suffices to track the user's evolving intent trajectory across topics, unresolved needs, and interest shifts. We propose OnePred, which maintains a recursively updated memory as its sole cross-turn context, bounding the per-turn cost inde
arXiv:2605.15607v2 Announce Type: replace Abstract: Large language models (LLMs) achieve high pass rates on code generation benchmarks, yet whether they can transfer this ability to languages absent from pretraining remains poorly understood. We introduce PyLang, a minimal imperative language absent from all pretraining corpora, and evaluate frontier models zero-shot and fine-tuned Qwen3 (4B, 8B, 32B) on 352 problems. We find that fine-tuning quickly teaches syntax but fails to transfer semantic competence: Python outperforms PyLang by up to 19% across all configurations, and no intervention (multi-task learning, preference tuning, code infilling, or latent-space objectives) closes the gap. An LLM judge reveals that frontier models select an identical algorithm to Python 80% of the time, yet cannot translate it into a working PyLang implementation., and CKA analysis confirms that fine-tuned models converge to nearly identical internal representations across languages (CKA > 0.97) while
arXiv:2605.08696v4 Announce Type: replace Abstract: Over the last two decades, language modeling has experienced a shift from the use of predominantly recurrent architectures that process tokens sequentially during training and inference to non-recurrent models that process sequence elements in parallel during training, which results in greater training efficiency and stability at the expense of lower inference throughput. Here we introduce the Structured Recurrent Mixer, an architecture that allows for algebraic conversion between a sequence parallel representation at train time and a recurrent representation at inference, notably without the need for specialized kernels or device-specific memory management. We show experimentally that this dual representation allows for greater training efficiency, higher input information capacity, and larger inference throughput and concurrency when compared to other linear complexity models. We postulate that recurrent models are poorly suited to
arXiv:2605.07725v2 Announce Type: replace Abstract: Tool-integrated reasoning (TIR) is difficult to scale to small language models due to instability in long-horizon tool interactions and limited model capacity. While reinforcement learning methods like group relative policy optimization provide only sparse outcome-level rewards. Recently, on-policy distillation (OPD) has gained popularity by supplying dense token-level supervision from a teacher on student-generated trajectories. However, our experiments indicate that applying OPD to TIR leads to a critical failure mode: erroneous tool calls tend to cascade across subsequent reasoning steps, progressively amplifying student-teacher divergence and rendering the teacher's token-level supervision increasingly unreliable. To address this, we propose SOD, a step-wise on-policy distillation framework for small language model agents, which adaptively reweights distillation strength at each step based on step-level divergence. Therefore, SOD
arXiv:2605.01350v2 Announce Type: replace Abstract: Detecting machine-generated text is essential for transparency and accountability when deploying LLMs. Watermarking enables statistically reliable detection by biasing token distributions to embed detectable signals into LLM outputs. However, it has been reported that watermarked LLMs often perform worse on downstream tasks. We propose PUPPET, a framework that fine-tunes an LLM via DPO to generate text that is both more detectable by a target detector and better performing on downstream tasks. We use two rewards: a detector that outputs a machine-class likelihood and an evaluator that measures a task-specific metric. Just as a watermark is verified with its secret key, this detector-specific design lets an LLM provider track how its published model is used. Experiments on long-form QA, summarization, and essay writing show that LLMs trained with PUPPET achieve detectability competitive with watermarking methods---even at strict low FP
arXiv:2605.00817v4 Announce Type: replace Abstract: Large language models (LLMs) often achieve strong performance on reasoning benchmarks, but final-answer accuracy alone does not show whether they faithfully execute the procedure specified in a prompt. We introduce a controlled diagnostic benchmark for arithmetic procedural execution, where models are given a step-wise arithmetic procedure and two numeric inputs, and must return the final computed value. Complexity is varied through procedure length and look-back dependencies over intermediate variables. Average first-answer accuracy drops from 63% on 5-step procedures to 20\% on 95-step procedures. Generation-level analysis shows that failures often involve missing answers, premature answers, self-correction after an initial error and under-executed traces. These findings reveal a consistent decline in execution performance as arithmetic procedural complexity increases.
arXiv:2604.10697v2 Announce Type: replace Abstract: Large language models frequently exhibit hallucinations: fluent and confident outputs that are factually incorrect or unsupported by the input context. While recent hallucination detection methods have explored various features derived from attention maps, the underlying mechanisms they exploit remain poorly understood. In this work, we propose SinkProbe, a hallucination detection method grounded in the observation that hallucinations are deeply entangled with attention sinks - tokens that accumulate disproportionate attention mass during generation - indicating a transition from distributed, input-grounded attention to compressed, prior-dominated computation. Importantly, although sink scores are computed solely from attention maps, we find that the classifier preferentially relies on sinks whose associated value vectors have large norms. Moreover, we show that previous methods implicitly depend on attention sinks by establishing the
arXiv:2604.06543v2 Announce Type: replace Abstract: In this work, we demonstrate that reliable stochastic sampling is a fundamental yet unfulfilled requirement for Large Language Models (LLMs) operating as agents. Agentic systems are frequently required to sample from distributions, often inferred from observed data, a process which needs to be emulated by the LLM. This leads to a distinct failure point: while standard RL agents rely on external sampling mechanisms, LLMs fail to map their internal probability estimates to their stochastic outputs. Through rigorous empirical analysis across multiple model families, model sizes, prompting styles, and distributions, we demonstrate the extent of this failure. Crucially, we show that while powerful frontier models can convert provided random seeds to target distributions, their ability to sample directly from specific distributions is fundamentally flawed.
arXiv:2604.03532v2 Announce Type: replace Abstract: Large language models (LLMs) show strong multilingual capabilities, yet reliably controlling the language of their outputs remains difficult. Representation-level steering addresses this by adding language-specific vectors to model activations at inference time, but identifying language-specific directions in the residual stream often relies on multilingual or parallel data that can be expensive to obtain. Sparse autoencoders (SAEs) decompose residual activations into interpretable, sparse feature directions and offer a natural basis for this search, yet existing SAE-based approaches face the same data constraint. We introduce LangFIR (Language Feature Identification via Random-token Filtering), a method that discovers language-specific SAE features using only a small amount of monolingual data and random-token sequences. Many SAE features consistently activated by target-language inputs do not encode language identity. Random-token s
arXiv:2604.03422v2 Announce Type: replace Abstract: All languages have a noun category, but its realisation varies considerably. Depending on the language, semantic and/or morphosyntactic differences may be more or less pronounced. This paper explores these variations, using Riffian as a reference point before extending the analysis to other languages. We propose a formal model termed morphology-driven marking. Nouns are organised into modular cognitive sets, each with its own morphological template and unmarked form. This approach helps explain differences in marking among noun types within and across languages. By situating these patterns within syntactic functions, we also reassess the notions of markedness and state. It is proposed that the concept of state be extended to all synthetic languages and analysed a novel subcategory of syntax-based inflection like agreement and grammatical case.
arXiv:2604.03136v5 Announce Type: replace Abstract: As AI-generated fiction becomes increasingly prevalent, questions of authorship and originality are becoming central to how written work is evaluated. While most existing work in this space focuses on identifying surface-level signatures of AI writing, we ask instead whether AI-generated stories can be distinguished from human ones without relying on stylistic signals, focusing on discourse-level narrative choices such as character agency and chronological discontinuity. We propose StoryScope, a pipeline that automatically induces a fine-grained, interpretable feature space of discourse-level narrative features across 10 dimensions. We apply StoryScope to a parallel corpus of 10,272 writing prompts, each written by a human author and five LLMs, yielding 61,608 stories, each ~5,000 words, and 304 extracted features per story. Narrative features alone achieve 93.2% macro-F1 for human vs. AI detection and 68.4% macro-F1 for six-way autho
arXiv:2603.28488v3 Announce Type: replace Abstract: Large language models (LLMs) remain unreliable for high-stakes claim verification due to hallucinations and shallow reasoning. While retrieval-augmented generation (RAG) and multi-agent debate (MAD) address this, they are limited by one-pass retrieval and unstructured debate dynamics. We propose a courtroom-style multi-agent framework, PROClaim, that reformulates verification as a structured, adversarial deliberation. Our approach integrates specialized roles (e.g., Plaintiff, Defense, Judge) with Progressive RAG (P-RAG) to dynamically expand and refine the evidence pool during the debate. Furthermore, we employ evidence negotiation, self-reflection, and heterogeneous multi-judge aggregation to enforce calibration, robustness, and diversity. In zero-shot evaluations on the Check-COVID benchmark, PROClaim achieves 81.7% accuracy, outperforming standard multi-agent debate by 10.0 percentage points, with P-RAG driving the primary perform
arXiv:2603.19258v2 Announce Type: replace Abstract: Differentially private (DP) fine-tuning of large language models (LLMs) requires massive compute and full model access, which rules out state-of-the-art proprietary APIs for general users. Generating DP synthetic data offers a practical workaround. This approach also allows for transparent exploratory data analysis and arbitrary reuse across downstream tasks, sidestepping the rigid constraints of a model's parameter space. Private Evolution (PE) provides a promising API-based framework for generating this data, but its success relies heavily on initialization. If the private data distribution falls too far outside the foundation model's pre-training priors -- a common issue in highly specialized domain -- PE struggles to align with the target data. This misalignment causes poor convergence, degraded utility, and wasted API calls. To solve this initialization bottleneck, we introduce Metadata Augmented Private Language Evolution (MAPLE
arXiv:2603.08091v2 Announce Type: replace Abstract: Large language model (LLM)-based judges are widely adopted for automated evaluation and reward modeling, yet their judgments are often affected by judgment biases. Accurately evaluating these biases is essential for ensuring the reliability of LLM-based judges. However, existing studies typically investigate limited biases under a single judge formulation, either generative or discriminative, lacking a comprehensive evaluation. To bridge this gap, we propose JudgeBiasBench, a benchmark for systematically quantifying biases in LLM-based judges. JudgeBiasBench defines a taxonomy of judgment biases across 4 dimensions, and constructs bias-augmented evaluation instances through a controlled bias injection pipeline, covering 12 representative bias types. We conduct extensive experiments across both generative and discriminative judges, revealing that current judges exhibit significant and diverse bias patterns that often compromise the rel
arXiv:2603.07475v4 Announce Type: replace Abstract: Autoregressive (AR) language models build representations incrementally via left-to-right prediction, while diffusion language models (dLLMs) are trained through full-sequence denoising. Although recent dLLMs match AR performance, whether diffusion objectives fundamentally reshape internal representations remains unclear. We perform the first layer- and token-wise representational analysis comparing native dLLMs (LLaDA), native AR models (Qwen2.5), and AR-initialized dLLMs (Dream-7B), using cosine similarity across layers and tokens alongside static inference-time layer-skipping as an analytical probe of redundancy. We find that diffusion objectives produce more global representations with substantial early-layer redundancy and reduced recency bias, while AR objectives yield tightly coupled, locally structured representations. AR-initialized dLLMs retain AR-like dynamics despite diffusion training, revealing persistent initialization
arXiv:2603.03320v2 Announce Type: replace Abstract: Effective communication often relies on aligning a message with an audience's narrative and worldview. Narrative shift involves transforming text to reflect a different narrative framework while preserving its original core message--a task we demonstrate is significantly challenging for current Large Language Models (LLMs). To address this, we propose a neurosymbolic approach grounded in social science theory and abductive reasoning. Our method automatically extracts rules to abduce the specific story elements needed to guide an LLM through a consistent and targeted narrative transformation. Across multiple LLMs, abduction-guided transformed stories shifted the narrative while maintaining the fidelity with the original story. For example, with GPT-4o we outperform the zero-shot LLM baseline by 55.88% for collectivistic to individualistic narrative shift while maintaining superior semantic similarity with the original stories (40.4% im
arXiv:2603.00025v3 Announce Type: replace Abstract: Direct Preference Optimization (DPO) is effective for offline alignment but poorly matched to ontology-driven structured prediction, where preferred and rejected JSON often differ by only a few schema-defining tokens. In this low-edit-distance regime, sequence-level DPO spreads gradient mass across non-critical serialization tokens (gradient dilution) and can reduce likelihood on rare preferred schema tokens (token erosion). To address these limitations, we first develop a confusion-aware preference-construction strategy combining expert-curated ambiguity patterns with validation-set SFT structured-error modes, producing minimally perturbed, schema-valid negatives for ontology-level decision errors. We then introduce Token-Adaptive Barrier Preference Optimization (TAB-PO), a post-SFT objective for token-critical structured generation with a confidence-gated token-level barrier that anchors under-confident schema tokens. On SciERC, wit
arXiv:2602.08404v3 Announce Type: replace Abstract: Diffusion large language models (dLLMs) have recently gained significant attention due to their inherent support for parallel decoding. Building on this paradigm, Mixture-of-Experts (MoE) dLLMs with autoregressive (AR) initialization have further demonstrated strong performance competitive with mainstream AR models. However, we identify a fundamental mismatch between MoE architectures and diffusion-based decoding. Specifically, a large number of experts are activated at each denoising step, while only a small subset of tokens is ultimately accepted, resulting in substantial inference overhead and limiting their deployment in latency-sensitive applications. In this work, we propose TEAM, a plug-and-play framework that accelerates MoE dLLMs by enabling more accepted tokens with fewer activated experts. TEAM is motivated by the observation that expert routing decisions exhibit strong temporal consistency across denoising levels as well a
arXiv:2602.06423v2 Announce Type: replace Abstract: Humor is a commonly used and intricate human language in daily life. Humor generation, especially in multi-modal scenarios, is a challenging task for large language models (LLMs), which is typically as funny caption generation for images, requiring visual understanding, humor reasoning, creative imagination, and so on. Existing LLM-based approaches rely on reasoning chains or self-improvement, which suffer from limited creativity and interpretability. To address these bottlenecks, we develop a novel LLM-based humor generation mechanism based on a fundamental humor theory, GTVH. To produce funny and script-opposite captions, we introduce a humor-theory-driven multi-role LLM collaboration framework augmented with humor retrieval (HOMER). The framework consists of three LLM-based roles: (1) conflicting-script extractor that grounds humor in key script oppositions, forming the basis of caption generation; (2) retrieval-augmented hierarchi
arXiv:2602.05220v4 Announce Type: replace Abstract: Current audio foundation models typically rely on rigid, task-specific supervision (e.g., speech recognition), addressing isolated factors of audio rather than the whole. In contrast, human processes audio holistically, seamlessly bridging raw audio waveform with abstract cognitive concepts (e.g., all perception details of audio events) to execute complex tasks. Grounded in this philosophy, we introduce Bagpiper, an 8B audio foundation model that interprets physical audio via rich captions, i.e., comprehensive natural language descriptions that encapsulate the critical cognitive concepts inherent in the audio. By pre-training on a massive corpus of 600B tokens, the model establishes a robust bidirectional mapping between raw audio and this high-level conceptual space. During fine-tuning, Bagpiper adopts a caption-then-process workflow, simulating an intermediate cognitive reasoning step to solve diverse tasks without knowing prior tas
arXiv:2602.03980v2 Announce Type: replace Abstract: Any language model must decide what to say in novel contexts based on information from similar contexts. But what about contexts that are not novel but merely infrequent? In hierarchical regression, the model's predictions for behavior in a context are affected by observations from similar contexts to the extent that 1) the current context is infrequent and 2) different contexts behave similarly. This is called adaptive partial pooling. This paper shows that next-word predictions of a transformer (GPT2) are affected by observations from outside the current context, but this pooling reduces with more training. Pooling is affected by context frequency, context number (type frequency) and context variability in a qualitatively similar way to hierarchical regression. However, there is a "sweet spot" in training at which the transformer best matches the behavior of hierarchical regression. This is the point at which the effect of context f
arXiv:2602.01132v2 Announce Type: replace Abstract: Tasks such as solving arithmetic equations, evaluating truth tables, and completing syllogisms are handled well by large language models (LLMs) in their standard form, but they often fail when the same problems are posed in logically equivalent yet obfuscated formats. To study this vulnerability, we introduce Logifus, a structure-preserving logical obfuscation framework, and, utilizing this, we present LogiQAte, a first-of-its-kind diagnostic benchmark with 1,108 questions across four reasoning tasks: (i) Obfus FOL (first-order logic entailment under equivalence-preserving rewrites), (ii) Obfus Blood Relation (family-graph entailment under indirect relational chains), (iii) Obfus Number Series (pattern induction under symbolic substitutions), and (iv) Obfus Direction Sense (navigation reasoning under altered directions and reference frames). Across all the tasks, evaluating six state-of-the-art models, we find that obfuscation severel
arXiv:2602.00740v2 Announce Type: replace Abstract: Medical text editing is essential for improving communication among diverse stakeholders in clinical settings. However, adapting LLM agents to this task remains challenging because expert supervision is often sparse, fragmented, and distributed across interacting quality dimensions. We identify that direct accumulation or retrieval of individual feedback is insufficient for effective adaptation, as fragmented evaluations do not directly translate into a coherent understanding of medical text quality. Based on this observation, we propose MedTextWeaver, a training-free framework that transforms fragmented evaluative evidence into global quality principles and actionable procedural knowledge for medical text editing. Across three clinical text datasets and a real-world validation experiment, MedTextWeaver consistently improves performance over strong LLM baselines and existing memory-based adaptation approaches. Further analysis demonst
arXiv:2602.00665v4 Announce Type: replace Abstract: Customer-service question answering (QA) systems increasingly rely on conversational language understanding. While Large Language Models (LLMs) achieve strong performance, their high computational cost and deployment constraints limit practical use in resource-constrained environments. Small Language Models (SLMs) provide a more efficient alternative, yet their effectiveness for multi-turn customer-service QA remains underexplored, particularly in scenarios requiring dialogue continuity and contextual understanding. This study investigates instruction-tuned SLMs for context-summarized multi-turn customer-service QA, using a history summarization strategy to preserve essential conversational state. We also introduce a conversation stage-based qualitative analysis to evaluate model behavior across different phases of customer-service interactions. Nine instruction-tuned low-parameterized SLMs are evaluated against three commercial LLMs
arXiv:2601.19913v4 Announce Type: replace Abstract: LLMs now help students plan, draft, and revise essays. Educational assessment therefore faces a basic question: how should student and LLM writing be compared? Rubrics assign points to content, organization, and expression. Their total can still hide which criteria drive the comparison, where ratings approach the maximum, and where readers disagree. We therefore conducted a secondary, post hoc audit of a Korean writing study with source-informed scoring. Three Korean language and literature majors used a 16-criterion, 100-point rubric to score six student essays and 24 essays from four LLMs prompted for three school levels. After source disclosure, they helped develop the rubric and, according to the protocol, scored the Phase 2 essays without gold labels while recording source judgments. The pooled LLM mean exceeds the student mean by 18.35 points. Orthographic norms and genre-appropriate register contribute 8.19 points, or 44.7% of
arXiv:2601.17753v2 Announce Type: replace Abstract: Research keyloggers are essential for cognitive studies of text production, yet most fail to capture the on-screen transformations performed by Input Method Editors (IMEs) for non-alphabetic scripts. To address this methodological gap, we present Hylog, a novel hybrid logging system that combines analytical keylogging with ecological text logging for a more complete and finer-grained analysis. Our modular, open-source system uses plug-ins for standard applications (Microsoft Word, Google Chrome) to capture both keyboard output and rendered text, which a hybridizer module then synchronizes into a dual trace. To validate the system's technical feasibility and demonstrate its analytical capabilities, we conducted a proof-of-concept study where two volunteers translated a text into simplified Chinese. Hylog successfully captured keypresses and temporal intervals between Latin letters, Chinese characters, and IME confirmations -- some meas
arXiv:2601.16800v4 Announce Type: replace Abstract: Fine-grained opinion analysis of text provides a detailed understanding of expressed sentiments and their targets. Although this level of detail is valuable, annotating opinions in datasets for model training requires considerable human effort and substantial cost, especially across diverse domains and real-world applications. To address this shortage of domain-specific labelled datasets, we explore the feasibility of LLMs as automatic annotators for fine-grained opinion analysis. We use a declarative annotation pipeline, an approach that reduces the variability of manual prompt engineering when using LLMs to identify fine-grained opinion spans in text. We also present a dedicated methodology for an LLM to adjudicate multiple labels and produce final annotations, benchmarked against exact, flexible, and element-wise variants of a rule-based voting aggregator. We trial the pipeline with models of different sizes for the Aspect Sentimen
arXiv:2601.05713v2 Announce Type: replace Abstract: Understanding how large language models (LLMs) represent natural language is a central challenge in natural language processing (NLP) research. Many existing methods extract word embeddings from an LLM, visualise the embedding space via point-plots, and compare the relative positions of certain words. However, this approach only considers single words and not whole natural language expressions, thus disregards the context in which a word is used. Here we present a novel tool for analysing and visualising information flow in natural language expressions by applying diffusion tensor imaging (DTI) to word embeddings. We find that DTI reveals how embedding space representations change between tokens. Tracking these changes within the layers of an LLM allows for comparing different model structures and could potentially reveal opportunities for pruning an LLM's under-utilised layers. Our results show that our visualisation method permits n
arXiv:2512.10739v3 Announce Type: replace Abstract: Large Reasoning Models (LRMs) have expanded the mathematical reasoning frontier through Chain-of-Thought (CoT) techniques and Reinforcement Learning with Verifiable Rewards (RLVR), capable of solving AIME-level problems. However, the performance of LRMs is heavily dependent on the extended reasoning context length. For solving ultra-hard problems like those in the International Mathematical Olympiad (IMO), the required reasoning complexity surpasses the space that an LRM can explore in a single round. Previous works attempt to extend the reasoning context of LRMs but remain prompt-based and built upon proprietary models, lacking systematic structures and training pipelines. Therefore, this paper introduces Intern-S1-MO, a long-horizon math agent that conducts multi-round hierarchical reasoning, composed of an LRM-based multi-agent system including reasoning, summary, and verification. By maintaining a compact memory in the form of lem
arXiv:2511.21759v2 Announce Type: replace Abstract: Diffusion-based large language models (dLLMs) have recently gained significant attention for their exceptional performance and inherent potential for parallel decoding. Existing frameworks further enhance its inference efficiency by enabling KV caching. However, its bidirectional attention mechanism necessitates periodic cache refreshes that interleave prefill and decoding phases, both contributing substantial inference cost and constraining achievable speedup. Inspired by the heterogeneous arithmetic intensity of the prefill and decoding phases, we propose ODB-dLLM, a framework that orchestrates dual-boundaries to accelerate dLLM inference. In the prefill phase, we find that the predefined fixed response length introduces heavy yet redundant computational overhead, which affects efficiency. To alleviate this, ODB-dLLM incorporates an adaptive length prediction mechanism that progressively reduces prefill overhead and unnecessary comp
arXiv:2511.21692v2 Announce Type: replace Abstract: We investigate how well large language models (LLMs) generalize across different task difficulties, a key question for effective data curation and evaluation. Existing research is mixed regarding whether training on easier or harder data leads to better results, and whether those gains come on easier or harder test data. We address this question by conducting a systematic evaluation of LLMs' generalization across models, datasets, and fine-grained groups of example difficulty. We rank examples in six datasets using the outputs of thousands of different LLMs and Item Response Theory (IRT), a well-established difficulty metric in educational testing. Unlike prior work, our difficulty ratings are therefore determined solely by the abilities of many different LLMs, excluding human opinions of difficulty. With a more objective, larger-scale, and finer-grained analysis, we show that cross-difficulty generalization is often limited; training
arXiv:2511.20639v4 Announce Type: replace Abstract: Multi-agent systems (MAS) extend large language models (LLMs) from independent single-model reasoning to coordinative system-level intelligence. While existing LLM agents depend on text-based mediation for reasoning and communication, we take a step forward by enabling models to collaborate directly within the continuous latent space. We introduce LatentMAS, an end-to-end training-free framework that enables pure latent collaboration among LLM agents. In LatentMAS, each agent first performs auto-regressive latent thoughts generation through last-layer hidden embeddings instead of text. Then, a shared latent working memory preserves and transfers each agent's internal representations and latent thoughts, ensuring lossless information exchange without re-encoding. We provide detailed theoretical analyses showing that LatentMAS achieves higher expressiveness and lossless information preservation with lower overall complexity than standar
arXiv:2510.20449v2 Announce Type: replace Abstract: Instruction tuning is crucial for aligning Large Language Models (LLMs), yet the quality of instruction-following data varies significantly. While high-quality data is paramount, it is often scarce; conversely, abundant low-quality data is frequently discarded, leading to substantial information loss. Existing data augmentation methods struggle to augment this low-quality data effectively, and the evaluation of such techniques remains poorly defined. To address this, we formally define the task of Instruction Distillation: distilling multiple low-quality and redundant inputs into high-quality and coherent instruction-output pairs. Specifically, we introduce a comprehensive data construction pipeline to create MIXTURE, a 144K-sample dataset pairing low-quality or semantically redundant imperfect instruction clusters with their high-quality distillations. We then introduce LM-Mixup, by first performing supervised fine-tuning on MIXTURE
arXiv:2510.09935v2 Announce Type: replace Abstract: While memes are often humorous, they are frequently used to disseminate hate, causing serious harm to individuals and society. Current approaches to hateful meme detection mainly rely on pre-trained language models. However, less focus has been dedicated to \textit{what make a meme hateful}. Drawing on insights from philosophy and psychology, we argue that hateful memes are characterized by two essential features: a \textbf{presupposed context} and the expression of \textbf{false claims}. To capture presupposed context, we develop \textbf{PCM} for modeling contextual information across modalities. To detect false claims, we introduce the \textbf{FACT} module, which integrates external knowledge and harnesses cross-modal reference graphs. By combining PCM and FACT, we introduce \textbf{\textsf{SHIELD}}, a hateful meme detection framework designed to capture the fundamental nature of hate. Extensive experiments show that SHIELD outperfo
arXiv:2510.03519v2 Announce Type: replace Abstract: Time series reasoning is crucial to decision-making in diverse domains, including finance, energy, and scientific discovery. While existing time series foundation models (TSFMs) can capture low-level dynamic patterns and provide accurate forecasting, further analysis usually requires additional background knowledge and sophisticated reasoning, which are lacking in most TSFMs but can be achieved through Large Language Models (LLMs). On the other hand, without expensive post-training, LLMs often struggle with the numerical understanding of time series data. Although it is intuitive to integrate the two types of models, developing effective training recipes that align the two modalities for reasoning tasks is still an open challenge. To this end, we propose TS-Reasoner that aligns the latent representations of TSFMs with the textual inputs of LLMs for downstream understanding/reasoning tasks. Specifically, we propose a simple yet effecti
arXiv:2510.01925v3 Announce Type: replace Abstract: Reward models (RMs) play a critical role in enhancing the reasoning performance of LLMs. For example, they can provide training signals to finetune LLMs during reinforcement learning (RL) and help select the best answer from multiple candidates during inference. In this paper, we provide a systematic introduction to RMs, along with a comprehensive survey of their applications in LLM reasoning. We first review fundamental concepts of RMs, including their architectures, training methodologies, and evaluation techniques. Then, we explore their key applications: (1) guiding generation and selecting optimal outputs during LLM inference, (2) facilitating data synthesis and iterative self-improvement for LLMs, and (3) providing training signals in RL-based finetuning. Finally, we discuss critical open questions regarding the selection, generalization, evaluation, and enhancement of RMs, based on existing research and our own empirical findin
arXiv:2509.24560v2 Announce Type: replace Abstract: Extended Chain-of-Thought (CoT) reasoning has significantly bolstered the capabilities of medical large language models (LLMs). However, current models exhibit static computational expenditure, applying lengthy reasoning processes indiscriminately to both simple queries and complex diagnostic cases. This inefficiency is particularly prohibitive in real-world healthcare, where clinical scenarios range from time-sensitive emergencies requiring rapid response to intricate pathologies demanding deep analysis. To address this, we propose \textbf{AdaThink-Med}\footnote{Code available at https://github.com/shaohao011/AdaThinkMed}, an end-to-end framework for adaptive reasoning via uncertainty-guided length calibration. Although the underlying mechanism is generalizable, we demonstrate its critical value in the medical domain, where balancing inference latency with diagnostic precision is paramount. AdaThink-Med leverages entropy-based uncert
arXiv:2509.22536v5 Announce Type: replace Abstract: The immense computational cost of training Large Language Models (LLMs) presents a major barrier to innovation. While FP8 training offers a promising solution with significant theoretical efficiency gains, its widespread adoption has been hindered by the lack of a comprehensive, open-source training recipe. To bridge this gap, we introduce an end-to-end FP8 training recipe that seamlessly integrates continual pre-training and supervised fine-tuning. Our methodology employs a fine-grained, hybrid-granularity quantization strategy to maintain numerical fidelity while maximizing computational efficiency. Through extensive experiments, including the continue pre-training of models on a 160B-token corpus, we demonstrate that our recipe is not only remarkably stable but also essentially lossless, achieving performance on par with the BF16 baseline across a suite of reasoning benchmarks. Crucially, this is achieved with substantial efficienc
arXiv:2509.21079v2 Announce Type: replace Abstract: Foundation models have shown remarkable capabilities in various domains, but their performance on complex, multimodal engineering problems remains largely unexplored. We introduce SoM-1K, the first large-scale multimodal benchmark dataset dedicated to evaluating foundation models on problems in the strength of materials (SoM). The dataset, which contains 1,065 annotated SoM problems, mirrors real-world engineering tasks by including both textual problem statements and schematic diagrams. Due to the limited capabilities of current foundation models in understanding complicated visual information, we propose a novel prompting strategy called Descriptions of Images (DoI), which provides rigorous expert-generated text descriptions of the visual diagrams as the context. We evaluate eight representative foundation models, including both large language models (LLMs) and vision language models (VLMs). Our results show that current foundation
arXiv:2508.08645v3 Announce Type: replace Abstract: As multimodal large language models advance rapidly, the automation of mobile tasks has become increasingly feasible through the use of mobile-use agents that mimic human interactions from graphical user interfaces. To further enhance mobile-use agents, previous studies employ demonstration learning to improve mobile-use agents from human demonstrations. However, these methods focus solely on the explicit intention flows of humans (e.g., step sequences) while neglecting implicit intention flows (e.g., personal preferences), which makes it difficult to construct personalized mobile-use agents. In this work, to evaluate the Intention Alignment Rate between mobile-use agents and humans, we first collect MobileIAR, a dataset covering 40 English- and Chinese-speaking users and containing human-intent-aligned actions and ground-truth actions. This enables a comprehensive assessment of the agents' understanding of human intent. Then we propo