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:2508.02208v3 Announce Type: replace Abstract: Evaluating the mathematical capability of Large Language Models (LLMs) is a critical yet challenging frontier. Existing benchmarks fall short, particularly for proof-centric problems, as manual creation is unscalable and costly, leaving the true mathematical abilities of LLMs largely unassessed. To overcome these barriers, we propose Proof2Hybrid, the first fully automated framework that synthesizes high-quality, proof-centric benchmarks from natural language mathematical corpora. The key novelty of our solution is Proof2X, a roadmap of converting mathematical proofs into various kinds of questions that are easy to verify. Instructed by this roadmap, we propose a new type of hybrid-formatted questions, named ``$m$-out-of-$n$ multiple judge questions'', specifically designed to enable robust, automatic evaluation while being resilient to guessing and superficial pattern matching inherent in traditional formats. As a demonstration of ou
arXiv:2507.06415v3 Announce Type: replace Abstract: Long-context reasoning requires accurately identifying relevant information in extensive, noisy input contexts. In this work, we propose PERK (Parameter Efficient Reasoning over Knowledge), a scalable approach for learning to encode long contexts using gradient updates at test time. Specifically, PERK employs two nested optimization loops in a meta-training phase. The inner loop rapidly encodes contexts into a low-rank adapter (LoRA) that serves as a parameter-efficient memory module for the base model. Concurrently, the outer loop learns to use the updated adapter to accurately recall and reason over relevant information from the encoded long context. Our evaluations on several long-context reasoning tasks show that PERK significantly outperforms the standard long-context finetuning, achieving average absolute performance gains of up to 20% for Qwen-2.5 (0.5B & 7B) on synthetic and real-world long-context reasoning. PERK also maintai
arXiv:2505.22635v2 Announce Type: replace Abstract: A common approach for teaching large language models (LLMs) to reason is to train on chain-of-thought (CoT) traces of in-distribution reasoning problems, but such annotated data is costly to obtain for every problem of interest. We want reasoning models to generalize beyond their training distribution, and ideally to generalize compositionally: combine atomic reasoning skills to solve harder, unseen reasoning tasks. We take a step towards compositional generalization of reasoning skills when addressing a target compositional task that has no labeled CoT data. We find that simply training models on CoT data of atomic tasks leads to limited generalization, but minimally modifying CoT formats of constituent atomic tasks to be composable can lead to improvements. We can train "atomic CoT" models on the atomic tasks with Composable CoT data and combine them with multitask learning or model merging for better zero-shot performance on the ta
arXiv:2505.17537v2 Announce Type: replace Abstract: Large language models (LLMs) often produce incorrect answers with high confidence, yet the factors associated with such overconfidence remain insufficiently understood. We study this problem through the lens of knowledge popularity. Using entity-centric factual QA derived from Wikidata triplets, we characterize popularity through question entity popularity, answer popularity, and question-answer co-occurrence. We find two consistent patterns. First, hallucinated answers are far from random: compared with ground-truth answers, they tend to be more popular or more frequently associated with the question entity. Second, confidence is strongly tied to the popularity of generated answers: even among incorrect predictions, more popular answers or those with higher question-answer co-occurrence receive higher confidence. Together, these findings suggest that popular but wrong alternatives may contribute to overconfidence. We further show tha
arXiv:2505.16782v3 Announce Type: replace Abstract: Large Language Models (LLMs) have shown impressive performance on complex tasks through Chain-of-Thought (CoT) reasoning. However, conventional CoT relies on explicitly verbalized intermediate steps, which constrains its broader applicability, particularly in abstract reasoning tasks beyond language. To address this, there has been growing research interest in \textit{latent CoT reasoning}, where the reasoning process is embedded within latent spaces. By decoupling reasoning from explicit language generation, latent CoT offers the promise of richer cognitive representations and facilitates more flexible, faster inference. This paper aims to present a comprehensive overview of this emerging paradigm and establish a systematic taxonomy. We analyze recent advances in methods, categorizing them from token-wise horizontal approaches to layer-wise vertical strategies. We then provide in-depth discussions of these methods, highlighting their
arXiv:2505.12196v2 Announce Type: replace Abstract: The impressive linguistic abilities of large language models (LLMs) have recommended them as models of human sentence processing, with some conjecturing a positive 'quality-power' relationship, in which language models' (LMs') fit to psychometric data continues to improve as their ability to predict words in context increases. This is important because it might suggest that elements of LLM architecture reflect the architecture of the human sentence processing faculty, and that any inadequacies in predicting human reading time and brain imaging data may be attributed to insufficient model complexity, which recedes as larger models become available. But recent studies have shown this scaling inverts after a point, as LMs become excessively large and accurate, when information-theoretic surprisal is used as a predictor. Other studies propose the use of entire vectors from differently sized LLMs, still showing positive scaling, casting do
arXiv:2504.14707v4 Announce Type: replace Abstract: We introduce FLAME (FLemish Accounts of Momentary Experiences), a corpus of nearly 25,000 personal narratives in Belgian-Dutch (Flemish), collected through experience sampling to support Natural Language Processing (NLP) research on an underrepresented variety. Such everyday narratives are rich in culturally grounded themes, but their informal register and low-resource setting make thematic extraction hard. Comparing K-Means, LDA, and BERTopic, we find that human evaluation favors BERTopic, which produces the most coherent, culturally resonant topics. FLAME, thereby, offers a new resource for studying everyday language use in a low-resource variety.
arXiv:2503.04822v2 Announce Type: replace Abstract: Graph neural networks (GNNs) have demonstrated success in modeling relational data primarily under the assumption of homophily. However, many real-world graphs exhibit heterophily, where linked nodes belong to different categories or possess diverse attributes, such as webpages, Wikipedia articles, social networks, and e-commerce platforms. Additionally, nodes in many domains are associated with textual descriptions, forming heterophilic text-attributed graphs (TAGs). Despite their significance, heterophilic TAGs remain underexplored due to the lack of dedicated benchmarks that jointly capture heterophilic structures and rich textual attributes. To address this gap, we introduce the \textbf{He}terophilic \textbf{T}ext-attributed \textbf{G}raph \textbf{B}enchmark (HeTGB), a novel benchmark comprising five real-world heterophilic graph datasets from diverse domains, with nodes enriched by extensive textual descriptions. HeTGB enables sy
arXiv:2502.11439v3 Announce Type: replace Abstract: Fine-tuning is an important step in adapting foundation models such as large language models to downstream tasks. To make this step more accessible to users with limited computational budgets, it is crucial to develop fine-tuning methods that are memory and computationally efficient. Sparse Fine-tuning (SpFT) and Low-rank adaptation (LoRA) are two frameworks that have emerged for addressing this problem and have been adopted widely in practice. In this work, we develop a new SpFT framework, based on ideas from neural network pruning. At a high level, we first identify "important" neurons/nodes using feature importance metrics from network pruning (specifically, we use the structural pruning method), and then perform fine-tuning by restricting to weights involving these neurons. Experiments on common language tasks show our method improves SpFT's memory efficiency by 20-50\% while matching the accuracy of state-of-the-art methods like
arXiv:2502.10577v2 Announce Type: replace Abstract: Instruct-based large language models (LLMs) have been shown to propagate and even amplify gender bias when prompted with contextually constrained instructions (e.g., writing a text from a description or selecting a gendered pronoun). However, little attention has been paid to biases in responses to contextually unconstrained (generic) instructions conveyed by gendered language, particularly masculine generics (MG). MG, found in many gender-marked languages, denote the use of the masculine gender as a supposedly neutral reference to mixed-gender groups or individuals whose gender is unknown or non-binary. Yet, psycholinguistic studies demonstrate that MG are not neutral and systematically induce gender bias. This study investigates how both local and proprietary LLMs are MG-biased when responding to generic prompts in French, examining LLMs' MG bias rates and use of gender-fair language (GFL). We create a 16k+ human noun database from
arXiv:2502.00857v2 Announce Type: replace Abstract: Large Language Models (LLMs) increasingly provide direct answers to user questions, raising concerns about reduced engagement in critical thinking and problem-solving. Hint generation offers an alternative by guiding users toward answers without revealing them, while hint evaluation assesses the quality of such guidance. Research in this area is hindered by fragmented datasets, inconsistent annotation formats, and evaluation tools that are often dataset-specific or unavailable. To address these challenges, we introduce HintEval, an open-source Python library for unified hint generation and evaluation. HintEval standardizes access to diverse hint datasets, supports answer-aware and answer-agnostic generation methods, and implements multiple evaluation metrics within a shared data model. The toolkit enables reproducible experimentation, cross-dataset analysis, and multi-dimensional evaluation with minimal engineering effort. We further
arXiv:2412.18707v3 Announce Type: replace Abstract: Multiple valid translations of a single literary work naturally exist. We investigate strategies for leveraging these multi-reference datasets to improve literary machine translation. We propose a filtering framework based on semantic similarity to identify source texts whose references display meaningful variation while remaining faithful. We find that fine-tuning with medium to high semantic similarity data substantially outperforms low semantic similarity data. Moreover, using medium and high semantic similarity data achieves comparable or better performance than using the full unfiltered data. Synthetic translations generated by LLMs are economical and convenient alternatives to human expert translations; however, we find fine-tuning on human expert translations outperforms fine-tuning on synthetically augmented data in automatic metrics and human evaluations, demonstrating the indispensable value of human expert translations for
arXiv:2412.15957v2 Announce Type: replace Abstract: The rapid development of large language models (LLMs) has transformed many industries, including healthcare. In practice, hospitals and patients increasingly seek LLM-based systems capable of interpreting personal health records and providing healthcare guidance. However, existing approaches mainly rely on general medical knowledge and often fail to account for individual variability, limiting their ability to provide personalized guidance. To address this, we propose personalized prompt learning (PPL), a framework that learns individualized prompts to guide LLMs in generating personalized healthcare recommendations. PPL constructs initial personalized prompts by leveraging both self-informed patient information and peer-informed signals derived from clinically similar cases. These prompts are then refined using reinforcement learning (RL) to better align the generated responses with physician recommendations written for each patient.
arXiv:2405.18540v3 Announce Type: replace Abstract: Red-teaming, or identifying prompts that elicit harmful responses, is a critical step in ensuring the safe and responsible deployment of large language models (LLMs). Developing effective protection against many modes of attack prompts requires discovering diverse attacks. Automated red-teaming typically uses reinforcement learning to fine-tune an attacker language model to generate prompts that elicit undesirable responses from a target LLM, as measured, for example, by an auxiliary toxicity classifier. We show that even with explicit regularization to favor novelty and diversity, existing approaches suffer from mode collapse or fail to generate effective attacks. As a flexible and probabilistically principled alternative, we propose to use GFlowNet fine-tuning, followed by a secondary smoothing phase, to train the attacker model to generate diverse and effective attack prompts. We find that the attacks generated by our method are ef
arXiv:2311.13892v4 Announce Type: replace Abstract: The social biases and unwelcome stereotypes revealed by pretrained language models are becoming obstacles to their application. Compared to numerous debiasing methods targeting word level, there has been relatively less attention on biases present at phrase level, limiting the performance of debiasing in discipline domains. In this paper, we propose an automatic multi-token debiasing pipeline called \textbf{General Phrase Debiaser}, which is capable of mitigating phrase-level biases in masked language models. Specifically, our method consists of a \textit{phrase filter stage} that generates stereotypical phrases from Wikipedia pages as well as a \textit{model debias stage} that can debias models at the multi-token level to tackle bias challenges on phrases. The latter searches for prompts that trigger model's bias, and then uses them for debiasing. State-of-the-art results on standard datasets and metrics show that our approach can si
arXiv:2608.31105v1 Announce Type: cross Abstract: Users of a deployed language model routinely encounter behaviours that testing almost never surfaces, since deployment puts the model through orders of magnitude more interactions than any evaluation can simulate. Automated auditors make testing cheap to scale and flexible enough to cover almost any specified behaviour, yet their lack of optimisation pressure makes them sample-inefficient. To address this shortcoming, we introduce BLOOM-WILT, a full auditing pipeline that elicits natural multi-turn instances of rare behaviours, without training cost or access beyond the target's next-token distribution. On the input side, WILT's auditor model revises its conversational strategy across rounds, learning from previous scored interactions. On the output side, WILT adaptively reweights the target's decoding using the model's own distribution conditioned on an elicitation prompt, so that behaviour-relevant generations are sampled ahead of oth
arXiv:2608.31082v1 Announce Type: cross Abstract: Valuable data remains embedded in unstructured sources: web pages, reports, contracts, filings, earnings calls, and PDFs. The big bet in enterprise AI is deploying LLM agents that reason over this data to answer complex questions for every knowledge worker. Agents can do this today, but at prohibitive cost. Each question repeatedly opens large documents to recover scattered evidence, consuming up to a million tokens. However, if the data were already structured, the same question would reduce to a cheap database lookup. For example, on FanOutQA benchmark, reasoning over an ideal pre-structured store is 28X cheaper, and the gap grows to orders of magnitude as questions fan out over more documents. Yet structuring everything in advance is not viable: documents hold vastly more possible structure than any workload will use, and the useful structure and documents are unknown until queries arrive. We propose agentic data cracking, a method t
arXiv:2608.31069v1 Announce Type: cross Abstract: Large language models decode by projecting hidden states through a large vocabulary head at every step. This operation is computationally costly and forces all reasoning to be expressed in discrete tokens. We introduce Soft Latent Thinking, a method that replaces the LM head during reasoning with a lightweight projector, enabling autoregressive rollout in embedding space where reasoning steps remain continuous rather than tokenized. Experiments on DeepSeek-Qwen-1.5B and LLaMA-3.2-3B show that Soft Latent Thinking consistently improves pass@k across all k while reducing per-step compute during chain-of-thought. Our method achieves the highest pass@32 among all soft-thinking approaches, demonstrating that effective reasoning can be carried out in continuous space without discrete token generation.
arXiv:2608.31059v1 Announce Type: cross Abstract: When do text embeddings work as inputs to empirical analysis? Their use rests on an assumption: that we can trade text for its low-dimensional embedding, and lose little in doing so. I make that assumption precise under a generative model in which documents are mixtures of latent topics. I study two uses---clustering units in embedding space and controlling for high-dimensional text. A cluster of embeddings is a set of documents with similar topic mixtures; controlling for the embedding is equivalent to controlling for the topic mixture, so validity reduces to whether that mixture captures the confounding. In an application to 363 U.S. metropolitan areas, embedding-based clusters of LLM-generated economic descriptions recover interpretable economic archetypes and separate local employment dynamics more sharply than clustering on model residuals, or on a curated set of industry and demographic covariates.
arXiv:2608.31046v1 Announce Type: cross Abstract: On-policy distillation (OPD) offers dense token-level supervision as an alternative to the sparse outcome-level advantages of reinforcement learning with verifiable rewards (RLVR). However, the teacher scores student-generated trajectories that are inherently off-policy for it, so the reliability of its supervision, and hence the source of the student's improvement, remains unclear. We quantitatively analyze teacher supervision during OPD training and find substantial noise whose prevalence increases with teacher scale. Surprisingly, the student policy is insensitive to such noise, converging to comparable performance regardless of whether noisy supervision is retained or removed. Does OPD distill at all? By analyzing what drives its gains, we find that learning concentrates on low log-probability tokens, and using a single fixed negative advantage matches the performance of teacher-provided ones. This suggests that OPD works largely by
arXiv:2608.30986v1 Announce Type: cross Abstract: Activation steering has emerged as a lightweight approach for controlling model refusal at inference time. A growing line of research explores trainable rotations of activations to develop geometrically principled intervention mechanisms. However, existing techniques rely on auxiliary constructs, such as refusal vectors, to define these rotations. In our work, we develop a self-contained methodology for learning parameter-efficient rotational transformations based on Riemannian optimization. We empirically validate the proposed scheme, demonstrating its superiority in intervention efficiency. An extensive ablation study highlights the importance of key design choices in our method. Our results identify the proposed rotation-based steering scheme as a promising direction for more reliable control over the behavior of LLMs.
arXiv:2608.30952v1 Announce Type: cross Abstract: Chemistry questions often demand exact computation and database lookups that a language model cannot supply from its parameters, so it must reach for external tools. Tool use here is a three-part problem: select the right tool from a large pool, fill it with correctly typed arguments, and chain calls so that each consumes the outputs of the last. CheMatAgent, a previously published system, addresses this with hierarchical evolutionary MCTS: separate policy and execution models searching tool-call trees under two learned critics, one regressed partly onto GPT-assigned scores. We show that a single policy suffices. Our model interleaves reasoning, tool calls, and returns in one left-to-right generation, trained by a supervised warm-up and then outcome-level reinforcement learning against a programmatic reward read directly off the gold call chain, which leaves no learned critic and no judge in the training loop. On ChemToolBench multiple-
arXiv:2608.30912v1 Announce Type: cross Abstract: Artificial intelligence (AI) and natural language processing (NLP) are increasingly used to extract, integrate, and interpret biomedical knowledge relevant to cancer genomics, yet their translation into routine clinical oncology has been comparatively slow. The central challenge is not computational capability alone, but trustworthy integration into clinical workflows. This review examines how NLP and AI support the cancer genomics pipeline, from literature mining and automated variant interpretation to clinical trial matching, knowledge graph construction, and multimodal data integration. We identify four interrelated translational failure domains: evidence inconsistency, explainability and uncertainty, data governance and reproducibility, and interoperability. Rather than considering these challenges in isolation, we take a systems-level view, focusing on their interaction across the translational pathway. We propose a conceptual fram
arXiv:2608.30910v1 Announce Type: cross Abstract: Spectroscopic structure elucidation is central to molecular analysis, but recent Large Language Model (LLM)-based methods mostly formulate it as direct spectrum-to-SMILES generation. Although this paradigm can leverage paired spectral data, it does not explicitly model the analytical workflow used by spectroscopists, such as diagnostic peak interpretation, fragment reasoning, formula constraints, and chemical consistency checking. In this paper, we introduce S3C-LLM, a skill-guided and code-grounded agentic LLM for spectrum-to-structure elucidation. Rather than directly predicting a molecule, S3C-LLM retrieves modality-specific spectroscopy skills, executes analysis code to instantiate these skills on the input spectra, and integrates the resulting peak-level evidence and formula constraints before generating SMILES. Specifically, we contribute a self-evolving spectroscopy skill library, a thinking-augmented skill-code trajectory constr
arXiv:2608.30841v1 Announce Type: cross Abstract: Large language models can often generate plausible mathematical reasoning traces, but reliably identifying the correct solution among multiple candidates remains a key challenge. Existing test-time reasoning pipelines typically rely on text-based verifiers that re-read each generated solution, making verification an expensive component of inference. Prior work has shown, however, that LLMs often encode correctness-related signals in their internal representations, including awareness of when their own answers are likely to be wrong. Building on this observation, we introduce HSRM, a lightweight hidden-state reward model that verifies candidate solutions by directly reading the generator's internal representations rather than re-processing its text. HSRM extracts hidden states from a frozen generator at reasoning-step boundaries and uses a small Transformer encoder to rank candidates. It is trained from self-generated trajectories with o
arXiv:2608.30823v1 Announce Type: cross Abstract: The vocal music of each language carries a distinctive sonic identity, even without instrumental accompaniment. We ask whether these differences are measurable and traceable to specific phonemes. To tackle this question, we introduce phoneme-conditional analysis, which isolates the acoustic effect of typologically distinctive phonemes by comparing marker syllables against matched non-marker controls within the same song, holding singer, melody, and genre constant. Across nine typologically diverse languages and thousands of songs, we measure effects along five acoustic dimensions. Song-level profiles built from these effects identify the language of an unaccompanied vocal at 85.5% balanced accuracy in a nine-way classification with folds grouped by artist; whether the separability arises by accumulation of the phoneme-local effects themselves is left open. Our findings suggest that phonological structure leaves systematic and measurable
arXiv:2608.30748v1 Announce Type: cross Abstract: Existing jailbreak evaluations typically characterize robustness using a single attack success rate (ASR) measured in a default configuration (the vanilla state). However, user-LLM interactions can induce diverse operational states beyond the vanilla state. In this work, we find that jailbreak robustness is highly fragile to operational-state variation: even when the attack remains fixed, changing only an ordinary system prompt not designed to affect safety can dramatically alter attack success rates. We systematically investigate this phenomenon across seven aligned models and three representative jailbreak attacks, observing substantial differences in ASR between vanilla and non-vanilla operational states. In one case, ASR increases by up to 56 percentage points (2% to 58%) solely due to a change in operational state. Remarkably, these increases occur even for attacks originally designed and optimized under vanilla-state evaluation. W
arXiv:2608.30730v1 Announce Type: cross Abstract: Long-horizon agentic tasks go beyond chaining short tasks over more interaction turns. Their evolving dynamic environments and long-range dependencies require Large Language Models (LLMs) to continually explore, learn from experience, and adapt their policies over thousands of steps. We introduce E-Commerce Bench, the first open-source benchmark that integrates multi-round counterpart negotiation and dynamic events into a year-long business operation. Over a 365-day year, an LLM agent concurrently runs multiple online stores, researching the market, negotiating with suppliers to source inventory, optimizing sales strategies, fulfilling orders, handling returns, and managing cash flow to maximize its end-of-year total assets. To construct a realistic merchant-side operating environment, the product and supplier data are derived from a real e-commerce platform, while a year-long calendar of promotions, natural disasters, and supply-chain
arXiv:2608.30703v1 Announce Type: cross Abstract: Runtime guardrails are essential for reliable large language model (LLM) deployment, yet existing approaches typically rely on independent, external models that introduce additional inference cost, delayed safety signals, and a capacity mismatch with increasingly capable base models. To address these issues, we introduce SingProbe, a lightweight intrinsic runtime guard that directly reuses hidden states produced during LLM inference and operates alongside autoregressive decoding. Within a unified framework, SingProbe continuously predicts query intent, response safety, and hallucination risk at the token level with negligible additional guardrail inference overhead, offering a "free-lunch" solution. We further introduce SingStreamBench, a benchmark designed to assess whether streaming guardrails remain inactive on benign prefixes while promptly detecting emerging unsafe content. Extensive experiments show that SingProbe achieves competi
arXiv:2608.30686v1 Announce Type: cross Abstract: Coding agents are increasingly used for software engineering tasks, including bootstrapping projects from third-party repositories whose integrity cannot be assumed. Prior work on repository poisoning largely focuses on attacker-controlled injection and disguise, but developers also shape risk through everyday invocation choices: what task to delegate, how to phrase the request, and which skills or rules to supply. We term these user-side choices Prompt-Level Configurations (PLCs) and introduce CIPR (Coding In Poisoned Repos), the first benchmark that systematically varies PLCs in poisoned real-world repositories. CIPR comprises 1,920 instances across 20 repositories, four task types, three social-media-grounded prompt styles, and three skill/rule conditions, and measures attack success rate (ASR) and agent alert rate (AR) using automated runtime and trace-based oracles. Our evaluation reveals two key insights: (1) Vulnerability is high
arXiv:2608.30650v1 Announce Type: cross Abstract: LLM agents need to sustain goal-consistent reasoning across long multi-turn interactions under strict resource constraints. However, as the multi-turn context accumulates, it can destabilize the underlying LLM's internal representation of task-relevant information from earlier turns, blurring the boundary between constructive reasoning and representation drift. We formulate multi-turn reasoning as a hidden-state trajectory of the underlying LLM that is characterized via two complementary signals: temporal curvature that captures the directional consistency of turn-to-turn updates, and variance slope which measures the expansion or contraction of the exploration space. Across four tasks and three underlying LLMs, we observed that these geometric signals distinguish between correct and incorrect episodes prior to completion. We further decompose each episode into three-action chains formed from four actions (Read, Write, Respond, Transfer
arXiv:2608.30597v1 Announce Type: cross Abstract: Direct Preference Optimization (DPO) simplifies alignment through pairwise comparisons but assumes all observed preferences are reliable. Real data often violates this assumption, leading to reversed, weak, or ambiguous labels that cause harmful policy updates. To address this, we propose Posterior Label Correction DPO (PLC-DPO) to robustly optimize preferences by routing each pair's training signal as a clean, flip, or tie case. The key idea is to use the calibrated policy-reference margin as online evidence to take appropriate correction actions. This reframes noisy preference learning as actively correcting supervision direction and strength rather than merely filtering suspicious examples. Across 57 dataset-model-benchmark cells, PLC-DPO obtains the best mean win rate against DPO (60.5 vs. 55.5 for the next-best method). Injected-noise and tie stress tests, human disagreement analysis, and self-confirmation diagnostics further show
arXiv:2608.30517v1 Announce Type: cross Abstract: Benchmark saturation and data contamination increasingly obscure genuine scientific reasoning in frontier LLMs. We introduce \textsc{ScienceArena}, an olympiad-style benchmark from thirteen public science competitions in physics, chemistry, and biology, including IPhO and IChO 2025--2026, IBO 2023, USAPhO 2026, and USNCO 2025. Its open-ended, multi-step problems use process-credit rubrics, making faithful scoring difficult. We build ScienceArena through an expert-audited digitization pipeline that converts official exams, figures, solutions, and rubrics into structured items verified by olympiad medalists. To scale evaluation beyond costly human grading, we calibrate LLM-as-judge against medalist ground truth on archived answers from five models across IPhO and IChO; two strong judges stay within one point of expert total scores. Medalist notes show that failures often stem from visual grounding, structure fidelity, and global problem c
arXiv:2608.30510v1 Announce Type: cross Abstract: For provenance research and art market studies, auction catalogs are an essential resource to trace specific objects over time and space. While historical auction catalogs follow established domain conventions, their internal formatting remains highly variable, and their large-scale analysis is currently restricted by the lack of machine-readable representations of the auction lots. We propose a pipeline to automatically extract structured lot-level metadata from German Sales, a large database of historical auction and sales catalogs from the 19th and 20th centuries. Using a manually annotated test set of representative catalog pages, we evaluate Vision-Language Models (VLMs) under varying prompt strategies and constrained decoding frameworks. To reflect the practical constraints faced by cultural heritage institutions, including budget, compute resources, and data privacy requirements, we benchmark the methods across different deployme
arXiv:2608.30457v1 Announce Type: cross Abstract: Multimodal geometry reasoning requires VLMs to extract precise visual relations and preserve them through multi-step deduction. Existing free-form traces obscure the decisions that determine the answer, and trajectory-level reinforcement learning distributes a single terminal signal across the entire response. We introduce credit-addressable reasoning, in which the semantic units exposed during inference also define where learning compares alternatives and assigns credit. We instantiate this principle with Code-CoT, which retains the diagram, represents visual relations as line-addressable executable code, and organizes reasoning into typed events, and CE-GRPO, which selects event boundaries using structural priors and type-normalized entropy, samples complete continuations from shared prefixes, and converts outcome differences into localized advantages. Across nine geometry benchmarks, CE-GRPO achieves an average accuracy of 76.04, out
arXiv:2608.30429v1 Announce Type: cross Abstract: LLM-based agent systems increasingly adopt skill-based architectures to reduce repetitive reasoning costs and improve stable, efficient task execution. Recent studies propose self-evolving agents that autonomously generate, refine, and reuse skills from past experiences to enable continuous capability evolution. However, autonomous skill evolution introduces a new attack surface in which malicious capabilities are generated, stored, and reused as legitimate skills. In this paper, we define EvoSkill Injection as a threat model targeting the autonomous skill generation and evolution pipeline of self-evolving agents. We further propose SARGE (Red-teaming Autonomous Skill Generation and Evolution in self-evolving agents), a red-teaming framework for evaluating this threat model through iterative generation, escalation, and reinforcement interactions. To support our framework, we construct EvoSkillBench, a benchmark dataset of malicious inte
arXiv:2608.30362v1 Announce Type: cross Abstract: As LLM agents take real-world actions through tools, indirect prompt injection (IPI) has emerged as a serious threat. The standard metric, Attack Success Rate (ASR), counts whether an injection succeeds but ignores what the user notices in the agent's final response. Looking at successful injection traces, we find two distinct outcomes: the agent executes the injection while returning an otherwise normal response, or reports the injected action in its final response, giving the user a chance to notice. We call these covert and overt successes. From the user's perspective, we decompose ASR into the Covert Success Rate (CSR), counting successes leaving no trace in the final response, and the Overt Success Rate (OSR), counting successes the user can detect. To understand what drives the gap, we analyze successful trajectories and find that the agent's behavior after the injection separates covert from overt: covert traces hand control back
arXiv:2608.30322v1 Announce Type: cross Abstract: Professional agent tasks often depend on conventions that are absent from public corpora, yet benchmarks rarely control whether an agent has access to those conventions. We introduce a knowledge-gated task-construction protocol that separates a task instruction from a compact artefact containing private conventions, reference tables, and utility operators. Construction-time provenance, byte-identical task instructions across the provided- and withheld-artefact conditions, leak audits, and executable witnesses make dependence on the artefact explicit and testable. Across fifteen calibration tasks, one frontier agent configuration achieves a 68.0% pass rate with the artefact and 0% without it; on one task, a plausible but incorrect artefact also yields 0% across five trials. Deterministic solvers and rule corpora provide exact ground truth for structured tasks, while named criterion-level rubrics support outputs that cannot be checked by
arXiv:2608.30315v1 Announce Type: cross Abstract: Token embeddings are the basic representational units that connect discrete tokens with continuous computation in language models. Although modern language models learn embeddings from random initialization through gradient-based training, the dynamical mechanism by which meaningful embedding structures emerge remains unclear. In this work, we identify that the evolving embedding structures are closely related to token-conditioned label and contextual distributions, which we formalize as probability signatures. We observe a progressive learning process, which we term Context Staircase: embeddings learn the low-order statistic signatures of the data before the high-order ones. More specifically, we observe that early in training they align with the simplest, context-free signature linking a token to its label, and as training proceeds, they progressively reflect signatures involving more and more context tokens. We then analyze the gradi
arXiv:2608.30226v1 Announce Type: cross Abstract: Modular compression has enabled considerable parameter reduction in LLMs while preserving strong language understanding and downstream task accuracy. However, existing joint modular compression methods primarily rely on activation statistics, leaving loss-sensitivity information and its module-level characterization underexplored. We investigate addressing this gap with LaMoC, a loss-aware modular compression methodology that blends activation and Empirical Fisher statistics through gradient-error alignment. LaMoC improves joint compression by selecting compression statistics that better align local module reconstruction error with the downstream loss. Our contributions are three-fold: (1) We characterize the Empirical Fisher as a module-level loss-aware proxy that can be blended with the activation statistics required for compression. (2) We reformulate joint modular compression as a two-tiered optimization problem that minimizes modul
arXiv:2608.30213v1 Announce Type: cross Abstract: Text recognition, or extracting electronic text from document images, has been indispensable for knowledge retrieval tasks, such as retrieval-augmented generation (RAG). For Khmer, extracted text is subject to an extra word segmentation step, as Khmer does not use any visible word delimiters to denote word boundaries. Thus, a recognition-then-segmentation pipeline for Khmer requires two separate sequential models; this is not only error-prone but also adds significant latency for large-scale document processing. This paper proposes a novel joint Khmer text recognition and word segmentation framework in a unified model. The proposed model, using a connectionist-temporal-classification (CTC) decoder for fast, parallel decoding, can be instructed to recognize Khmer text with ($b=1$) and without ($b=0$) word segmentation. Experimental results on different benchmark datasets of different document modalities (document, scene, and handwritten
arXiv:2608.30181v1 Announce Type: cross Abstract: We introduce A.X K2, a 688B-parameter Mixture-of-Experts (MoE) language model trained from scratch as a high-performance foundation for \emph{agentic} applications. Trained on approximately 8.5T tokens---fewer than its predecessor, A.X K1---on a smaller but higher-quality mixture with substantially expanded agentic and software-engineering data, it nonetheless improves over A.X K1 across the board, by over 30 percentage points on some benchmarks, reflecting large gains in token efficiency. To support long contexts efficiently, we introduce Sparse Gated Attention (SGA), which combines sparse attention with gated attention, and adopt Gated Norm (GN) to stabilize large-scale training. SGA is trained natively at 128K through a \emph{sparse} indexer warmup that optimizes the indexer against its own sparse top-$k$ selection rather than the dense attention distribution, making adaptation markedly cheaper: each query reads only 2,048 positions,
arXiv:2608.30125v1 Announce Type: cross Abstract: Current video-to-music (V2M) models lack semantic control and fail to penalize instruction violations, largely due to their reliance on reconstruction objectives and the representational bottleneck of static cross-modal conditioning in Diffusion Autoregressive (DAR) architectures. To resolve this, we introduce VIBE, a novel text-and-video-to-music (T+V2M) generation model that leverages: (1) Conditioning Connection, a depth-wise cross-layer conditioning mechanism that dynamically bridges the planning and diffusion refinement heads and (2) a comprehensive reward modeling taxonomy, optimizing for both hard, verifiable constraints (e.g., tempo, key) and soft, subjective qualities (e.g., musicality, multimodal alignment) with a structured 5-stage training curriculum. Upon evaluation using audio-visual alignment, instruction following, and audio quality metrics, along with a subjective human evaluation study, we observe that VIBE demonstrate
arXiv:2608.30067v1 Announce Type: cross Abstract: How do LLM agents come to both understand environments they act in and master tasks set within them? Through controlled experiments combining world-model training (next-state prediction) and policy training (reward maximization), we investigate this question. We dissect the resulting models through their additive parameter updates. Geometrically, we find effective world-model updates are low-rank and share an input-feature subspace with policy updates while writing to nearly orthogonal output directions, whether trained separately or sequentially. However, we find that, in projection interventions, the sequential update induces more robustness than separate policy RL when removing the world model's leading input directions, suggesting that it has learned alternative input pathways. Behaviorally, we find the sequentially trained agent explores a wider range of states and actions. Based on this, we ask: does policy training preserve world
arXiv:2608.30044v1 Announce Type: cross Abstract: Language models are commonly compared by averaging scores across a benchmark list with equal weight. Such lists grow through publication outside an explicit measurement design, so equal weighting turns the density of published benchmarks into an implicit capability weight: densely benchmarked regions count repeatedly. We introduce Balance of Benchmarks (BoB), which embeds benchmark descriptions and assigns each benchmark an inverse-density semantic weight. Nearby entries share aggregate influence at a disclosed density scale. After equating heterogeneous scores onto a common latent scale, a residual field uses the same geometry to condition model rankings on a task query. The two components serve distinct empirical roles. On a snapshot of 586 models and 14 benchmarks, BoB predicts which models are unusually strong on a held-out task beyond their general ability, reaching a profile correlation of 0.462 compared with 0.049 under equal wei
arXiv:2608.30023v1 Announce Type: cross Abstract: Generative engines such as ChatGPT, Gemini, and Perplexity answer buyer questions directly and name a shortlist of brands inside the answer. Studying how brands enter or fail to enter that shortlist requires demand-side data: what buyers in a category ask, what information they need, and which sources they trust. Existing large persona corpora are built for training-data diversity and carry neither a staged search-intent label nor a preferred-sources field, so they cannot be joined to supply-side recommendation measurements. We built and validated PersonaGen-1M, a corpus of 1,031,732 synthetic buyer personas spanning 511 industry labels and 4 market contexts, carrying 19,416,821 structured behavioral attributes, 5,160,046 of them search queries. Each persona carries a single primary_intent label covering its query set (78.3% informational, 17.4% commercial, 4.3% transactional) and a preferred_sources field naming the source types that b
arXiv:2608.29974v1 Announce Type: cross Abstract: Detecting hallucinations in Large Vision-Language Models (LVLMs) requires both accurate span localization and well-calibrated confidence scores. Fine-tuned generative VLMs excel at identifying hallucinated text spans but suffer from overconfidence and high inference latency. Discriminative sequence taggers offer deterministic speed and superior calibration but exhibit conservative span recall. We present SpanCalib-VLM, a hybrid dual-system for the SHROOM-Visions Shared Task that combines a multimodal sequence tagger, consisting of XLM-RoBERTa-Large fused with a SigLIP vision encoder via cross-attention, with our fine-tuned generative VLM (Qwen3.5-4B-SHROOM-SFT). Through a Union-Calibrated Fusion strategy, candidate spans from the generative model are re-scored with calibrated probabilities from the sequence tagger. On the SHROOM-Visions English evaluation split, our ensemble achieves a Pearson calibration correlation of 0.41 and an over
arXiv:2608.29943v1 Announce Type: cross Abstract: Large language models (LLMs) can memorize sensitive information, raising serious privacy concerns. Machine unlearning offers a potential solution to remove such information, but it remains unclear whether existing methods truly erase it or merely hide it within the model. A key challenge is quantifying the persistence of sensitive data under a unified evaluation framework. To address this, we construct a synthetic dataset containing fake private information and propose a white-box auditing framework to systematically assess whether claimed-forgotten information is genuinely removed. Using this framework, we evaluate five existing unlearning methods and find that a simple "inverse greedy" decoding -- selecting the least likely token at each step -- can recover supposedly forgotten private information. Our results reveal that current unlearning approaches often fail to fully eliminate sensitive information, highlighting the need for more
arXiv:2608.29847v1 Announce Type: cross Abstract: Text-to-image models learn associations between concepts - in the case of this paper, people's professions, which we refer to as roles - and visual attributes. These associations can underpin many observed forms of stereotypical bias. A key open question in this area is whether these associations are stable or change when visual representations of people in professional roles are placed in different prompted contexts. We introduce ContextBias, a controlled evaluation framework, and ContextBench, a benchmark spanning 92 roles and 1,656 semantically controlled prompts, designed to isolate the effect of contextual variation on role-linked visual representations. Evaluating four state-of-the-art models on 66,240 generated images, we find that placing a role in a semantically unrelated context does not suppress role-linked attributes; instead, cross-role attribute concentration increases (pooled BI $+0.047$). Demographic cues, characteristic
arXiv:2608.29715v1 Announce Type: cross Abstract: Transformers rely on position embedding mechanisms in long context modeling in most cases. Rotary Position Embedding (RoPE) embeds positional information with independent 2D rotations, forming relative position terms in self-attention. However, its pairwise, block-based, and decoupled structure limits deep mixing and robustness across channels. We propose HD-RoPE, which extends RoPE from independent 2D rotations to higher-dimensional rotations and introduces a Paley-I orthogonal basis to obtain balanced, isotropic, and dense phase mixing within each rotation subspace. This significantly enhances channel coupling and rotational degrees of freedom while maintaining orthogonal stability and the relative position closure property. Furthermore, HD-RoPE is easily optimized for engineering efficiency without introducing additional trainable parameters. We have conducted extensive evaluation results demonstrating that HD-RoPE achieves significa