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.
Agios Pharmaceuticals said tebapivat fell short of expectations in a Phase 2 test in sickle cell disease. It’s the latest in a series of disappointing industry developments in sickle cell disease. The post Agios Pharma Drops Sickle Cell Disease Drug After Phase 2 Data Disappoint appeared first on MedCity News .
Screening for mental health symptoms in patients with significant allergic disease is not overreach The post How Diagnostic Precision in Allergy Care Supports Mental Well-Being appeared first on MedCity News .
Healthcare decision-makers, especially at pharmaceutical companies, must prioritize patient input. Data collected from social media listening is a significant step in the right direction. The post Why Pharma Companies Should Look To Social Media For Better Patient Insight appeared first on MedCity News .
A few years ago, Jenn Judkins was attending a data governance training session when the instructor posed a question: “You may say you have a good data privacy practice in place, but is it documented? If it’s not documented, does it really exist?” That question resonated with Judkins, technology director at Wayland Public Schools in Massachusetts. She had spent years thinking about how to protect student data, working with her team to follow best practices around access, data protection and vetting software vendors. But none of it was formally written down. The instructor, Melissa Tebbenkamp,…
When an AI system makes a “bad decision,” most organizations can’t really explain why. They aren’t entirely sure which data trained it, where the error was, which version it was running or which third-party components were involved. This is an ethical and regulatory blind spot that creates risk for tech leaders in institutions of all sizes, and it’s driving interest in a newer framework called the AI bill of materials. If you’ve ever seen a software bill of materials, or SBOM, then you already understand the concept. But many organizations have not yet thought about applying that framework…
[Sponsored] A five minute survey from MedCity News asks payers what they think of AI and how it is being implemented. Participate today and we will reward you! The post Where Will AI Deployment Benefit Payers the Most? appeared first on MedCity News .
Innovative Leader Award - Mary Decker of Franklin Special District offers 3 tips to help a middle school aerospace program take off
arXiv:2607.05690v2 Announce Type: replace-cross Abstract: Language agents run a loop - observe, reason, act - but the memory they reason over sits outside it: a store queried at most once per turn. We study the regime where memory moves inside the loop, read and written on every step. The obstacle has always been latency: networked stores answer in tens to hundreds of milliseconds, and in-loop retrieval can inflate end-to-end latency by up to 83x when retrieval is expensive. Prior work manages that cost rather than questioning it: serving-layer scheduling hides it, "memory-first" designs ration retrieval to once per turn. We argue latency is a property of where the store lives, not the in-loop pattern: an in-process store answers in ~100us, three orders of magnitude below the network regime, and at that speed the per-step tax collapses. By the extended-mind thesis's parity principle, a store fast enough to be constantly and directly available becomes extended working memory, not a tool
arXiv:2607.02947v2 Announce Type: replace-cross Abstract: Public official-information request records contain process signals. They can support research, workflow review, and analyst-led assessment. Yet they also mix observed correspondence, platform states, inferred events, and legal outcomes. FOI-O is a reusable process-modelling method and verification infrastructure for Freedom of Information administration. It is a global model that began with the New Zealand Official Information Act and has since iterated through the Australian Commonwealth and New South Wales settings. The NZ package remains the mature reference implementation; the Australian work remains provisional pending empirical evaluation and jurisdiction-specific legal validation. FOI-O models request records, observed correspondence, controlled vocabularies, provenance, review queues, release metadata, and bounded analysis rules. Legally meaningful outcomes require certification by an authorised decision-maker. Its type
arXiv:2606.20959v2 Announce Type: replace-cross Abstract: Language models may encode both outdated facts and their newer replacements. We introduce Parametric Temporal Conflict (PTC), where the newer fact is present and recoverable, but the default forward pass prefers the outdated one. We release a deterministically verified benchmark of 8,746 Wikidata position-holder transitions and evaluate four open-weight language models across three families. A date-prefix prompt recovers the newer fact in 61-81% of PTC cases. Activation patching flips predictions in 72-85% of cases and localizes the preference to model-specific upper-layer regions. Residual-stream steering outperforms norm-matched random directions, indicating direction-specific representations. These results show that PTC reflects a localized representational preference rather than missing knowledge. Recovery is measured on oracle-identified conflicts because automatic detection remains unreliable. We release the benchmark, cod
arXiv:2606.19341v2 Announce Type: replace-cross Abstract: Passive models for long video understanding typically rely on a "watch-it-all" paradigm, processing frames uniformly regardless of query difficulty, causing computational cost to grow with video duration. Although interactive frameworks have emerged, they often rely on global pre-scanning, and their context cost still scales with video length. We propose OmniAgent, the first native omni-modal agent that formulates video understanding as a POMDP-based iterative Observation-Thought-Action cycle. OmniAgent executes on-demand actions to selectively distill audio-visual cues into a persistent textual memory, effectively decoupling reasoning complexity from raw video duration. To operationalize this, we introduce (1) Agentic Supervised Fine-Tuning to bootstrap native active perception via best-of-N trajectory synthesis with dual-stage quality control, and (2) Agentic Reinforcement Learning with TAURA (Turn-aware Adaptive Uncertainty R
arXiv:2606.17799v2 Announce Type: replace-cross Abstract: Coding agents have become a major mode of software engineering, but the benchmarks we use to compare them were designed in a pre-agent era: they collapse model, harness, and environment into a single end-to-end score, typically computed against one reference solution, with no component-level signal for iteration. We argue that current coding benchmarks are misaligned with agentic software engineering. A coding agent in practice is not a model: it is a system harness -- a composite of models, harnesses, contexts, environments, and feedback signals, any one of which can move the benchmark score by margins comparable to those between adjacent model generations. We discuss three symptoms: (i) benchmark scores conflate the model with the rest of the harness; (ii) grading against a single reference solution penalises equally valid alternatives; and (iii) the absence of signal at the level of individual harness components makes the end
arXiv:2605.14066v2 Announce Type: replace-cross Abstract: Early-stage Parkinson's disease (EarlyPD) detection from speech is clinically meaningful yet underexplored, and published results are hard to compare because studies differ in datasets, languages, tasks, evaluation protocols, and EarlyPD definitions. To address this issue, we propose the first benchmark for speech-based EarlyPD detection, with a speaker-independent split designed for fair and replicable cross-method evaluation on researcher-accessible datasets. The benchmark covers three common speech tasks and evaluates methods under different training-resource settings. We also present multi-dimensional evaluation breakdowns by dataset, aggregation level, gender, and disease stage to support fine-grained comparisons and clinical adoption. Our results provide a replicable reference and actionable insights, encouraging the adoption of this publicly available benchmark to advance robust and clinically meaningful speech-based Earl
arXiv:2604.28061v2 Announce Type: replace-cross Abstract: Numerous metascience studies and other initiatives have begun to monitor the prevalence of open science practices when it is more important to understand the 'downstream' effects or impacts of open science. PLOS and DataSeer have developed a new LLM-based indicator to measure an important effect of open science: the reuse of research data. Our results show a data reuse rate of 43%, which is higher than established bibliometric techniques. We show that data reuse can be measured at scale using LLMs and generative artificial intelligence. The positive effects of research data sharing and reuse may currently be underestimated.
arXiv:2603.16011v3 Announce Type: replace-cross Abstract: Large language model (LLM) coding agents increasingly operate at the repository level, motivating benchmarks that evaluate their ability to optimize entire codebases under realistic constraints. Existing code benchmarks largely rely on synthetic tasks, binary correctness signals, or single-objective evaluation, limiting their ability to assess holistic optimization behavior. We introduce FormulaCode, a benchmark for evaluating agentic optimization on large, real-world codebases with fine-grained, multi-objective performance metrics. FormulaCode comprises 957 performance bottlenecks mined from scientific Python repositories on GitHub, each paired with expert-authored patches and, on average, 264.6 community-maintained performance workloads per task, enabling the holistic ability of LLM agents to optimize codebases under realistic correctness and performance constraints. Our evaluations reveal that repository-scale, multi-objectiv
arXiv:2602.21492v2 Announce Type: replace-cross Abstract: Reinforcement learning (RL) has become a central post-training paradigm for large language models (LLMs), but its performance is highly sensitive to the quality of training problems. This sensitivity stems from the non-stationarity of RL: rollouts are generated by an evolving policy, and learning is shaped by exploration and reward feedback, unlike supervised fine-tuning (SFT) with fixed trajectories. As a result, prior work often relies on manual curation or simple heuristic filters (e.g., accuracy), which can admit incorrect or low-utility problems. We propose GradAlign, a gradient-aligned data selection method for LLM reinforcement learning that uses a small, trusted validation set to prioritize training problems whose policy gradients align with validation gradients, yielding an adaptive curriculum. We evaluate GradAlign across three challenging data regimes: unreliable reward signals, distribution imbalance, and low-utility
arXiv:2512.10054v3 Announce Type: replace-cross Abstract: Autoregressive language models expose one causal token frontier, even when the requested document contains sections that could be developed concurrently. Existing parallel-generation systems arrange external branches around an otherwise unchanged model. We instead formulate model-intrinsic parallel generation: a single trained architecture owns multiple causal frontiers and produces one next-token distribution for each frontier in every synchronized decoding round. The Parallel Decoder Transformer (PDT) retains a frozen shared lower knowledge trunk and replaces the upper trunk with three independently parameterized physical decoder stacks. A prompt-time set planner produces three unordered continuous outlines, each hard-routed to one decoder as persistent Plan-KV memory, while a finite product-quantized notes bus carries block-delayed latent messages among the decoders. Autoregression is preserved within each lane; same-round la
arXiv:2511.16478v2 Announce Type: replace-cross Abstract: Music Recommender Systems (MRSs) have long relied on an information retrieval framing, where progress is measured mainly through accuracy on retrieval-oriented subtasks. While effective, this reductionist paradigm struggles to address the deeper question of what makes a good recommendation. Attempts to broaden evaluation, through user studies or fairness analyses, have had limited impact. The emergence of Large Language Models (LLMs) disrupts this framework: LLMs are generative rather than ranking-based, making standard accuracy metrics questionable. They also introduce challenges such as hallucinations, knowledge cutoffs, non-determinism, and opaque training data, rendering traditional train or test protocols difficult to interpret. At the same time, LLMs create new opportunities, enabling natural language (NL) interaction and even allowing models to act as evaluators. This work argues that the shift toward LLM-driven MRSs requ
arXiv:2509.25624v3 Announce Type: replace-cross Abstract: As LLMs advance into autonomous agents with tool-use capabilities, they introduce security challenges that extend beyond traditional content-based LLM safety concerns. This paper introduces Sequential Tool Attack Chaining (\STAC), a novel multi-turn attack framework that exploits agent tool use. \STAC chains together tool calls that each appear harmless in isolation but, when combined, collectively enable harmful operations that only become apparent at the final execution step. At the core of \STAC is an automated, closed-loop pipeline that synthesizes executable multi-step tool chains, validates them through in-environment execution, and reverse-engineers stealthy multi-turn prompts that reliably induce agents to execute the verified malicious sequence. Using this framework, we generate and systematically evaluate 483 \STAC cases, featuring 1,352 sets of user-agent-environment interactions and spanning diverse domains, tasks, a
arXiv:2509.00078v2 Announce Type: replace-cross Abstract: The emergence of large language models (LLMs) has transformed spoken dialog systems, yet the optimal architecture for real-time on-device voice agents remains an open question. While end-to-end approaches promise theoretical advantages, cascaded systems (CSs) continue to outperform them in language understanding tasks, despite being constrained by sequential processing latency. In this work, we introduce ChipChat, a novel low-latency CS that overcomes traditional bottlenecks through architectural innovations and streaming optimizations. Our system integrates streaming (a) conversational speech recognition with mixture-of-experts, (b) state-action augmented LLM, (c) text-to-speech synthesis, (d) neural vocoder, and (e) speaker modeling. Implemented using MLX, ChipChat achieves sub-second response latency on a Mac Studio without dedicated GPUs, while preserving user privacy through complete on-device processing. Our work shows tha
arXiv:2508.20083v2 Announce Type: replace-cross Abstract: Retrieval-Augmented Generation (RAG) has become a standard approach for improving the reliability of large language models (LLMs). Prior work demonstrates the vulnerability of RAG systems by misleading them into generating attacker-chosen outputs through poisoning the knowledge base. However, we observe that the effectiveness of these attacks is substantially undermined in the real-world deployment, where LLMs exhibit a strong self-correction ability (SCA). This ability is typically triggered by the mainstream configuration of LLMs, indicating a substantial gap between idealized research settings and practical scenarios. To address this issue, we systematically reflect on the limitations of prior RAG attacks and introduce DisarmRAG, a novel poisoning paradigm that focuses on the retriever, instead of the conventional approach of only poisoning the knowledge base. By compromising the retriever, our method can inject arbitrary ant
arXiv:2507.06445v2 Announce Type: replace-cross Abstract: Interpretability research often predicts model responses to targeted mechanistic interventions. But can we predict responses to unseen input data? We propose and demonstrate this alternate objective by using model internals to predict their out-of-distribution (OOD) behavior. We train hundreds of Transformers on simple synthetic tasks, where perfect in-distribution accuracy is compatible with multiple OOD generalization rules. We successfully use attention patterns -- observed only on in-distribution data -- to predict which rule each model follows on OOD data. Our experiments decouple the mechanistic faithfulness of our interpretation from its predictive value; ablations reveal such internal patterns can suppress rather than support the rule they predict, showing observational analysis can forecast behavior even when causal analysis fails to support a simple cause-effect link. Our findings are a proof-of-concept for a new inter
arXiv:2507.05515v3 Announce Type: replace-cross Abstract: Vision-language models (VLMs) are facing the challenges of understanding and following multimodal assembly instructions, particularly when fine-grained spatial reasoning and precise object state detection are required. In this work, we explore LEGO Co-builder, a hybrid benchmark combining real-world LEGO assembly logic with programmatically generated multimodal scenes. The dataset captures stepwise visual states and procedural instructions, allowing controlled evaluation of instruction-following, object detection, and state detection. We introduce a unified framework and assess leading VLMs such as GPT-4o, Gemini, and Qwen-VL, under zero-shot and fine-tuned settings. We also evaluated the framework using a reasoning-focused model, GLM-4.1-thinking. Our results show that while object detection achieved high performance (98.16% with fine-tuned InstructBLIP), fine-grained scene understanding and assembly state detection remain chal
arXiv:2506.04141v2 Announce Type: replace-cross Abstract: The sequential structure of videos poses a challenge to the ability of multimodal large language models (MLLMs) to locate multi-frame evidence and conduct multimodal reasoning. However, existing video benchmarks mainly focus on understanding tasks, which only require models to match frames mentioned in the question (hereafter referred to as "question frame") and perceive a few adjacent frames. To address this gap, we propose MMR-V: A Benchmark for Multimodal Deep Reasoning in Videos. The benchmark is characterized by the following features. (1) Long-range, multi-frame reasoning: Models are required to infer and analyze evidence frames that may be far from the question frame. (2) Beyond perception: Questions cannot be answered through direct perception alone but require reasoning over hidden information. (3) Reliability: All tasks are manually annotated, referencing extensive real-world user understanding to align with common per
arXiv:2505.20161v2 Announce Type: replace-cross Abstract: Effective generalization in language models depends critically on the diversity of their training data. Yet existing diversity metrics often fall short of this goal, relying on surface-level heuristics that are decoupled from model behavior. This motivates us to ask: What kind of diversity in training data actually drives generalization in language models -- and how can we measure and amplify it? Through large-scale empirical analyses spanning over 300 training runs, carefully controlled for data scale and quality, we show that data diversity can be a strong predictor of generalization in LLM reasoning -- as measured by average model performance on unseen out-of-distribution benchmarks. We introduce G-Vendi, a metric that quantifies diversity via the entropy of model-induced gradients. Despite using a small off-the-shelf proxy model for gradients, G-Vendi consistently outperforms alternative measures, achieving strong correlatio
arXiv:2502.04411v3 Announce Type: replace-cross Abstract: Model merging aggregates Large Language Models (LLMs) finetuned on different tasks into a stronger one. However, parameter conflicts between models leads to performance degradation in averaging. While model routing addresses this issue by selecting individual models during inference, it imposes excessive storage and compute costs, and fails to leverage the common knowledge from different models. In this work, we observe that different layers exhibit varying levels of parameter conflicts. Building on this insight, we average layers with minimal parameter conflicts and use a novel task-level expert routing for layers with significant conflicts. To further reduce storage costs, inspired by task arithmetic sparsity, we decouple multiple fine-tuned experts into a dense expert and several sparse experts. Considering the out-of-distribution samples, we select and merge appropriate experts based on the task uncertainty of the input data
arXiv:2412.18613v2 Announce Type: replace-cross Abstract: Illusions are entertaining, but they are also a useful diagnostic tool in cognitive science, philosophy, and neuroscience. A typical illusion shows a gap between how something `really is' and how something `appears to be', and this gap helps us understand the mental processing that led to how something appears to be. Illusions are also useful for investigating artificial systems, and much research has examined whether computational models of perception fall prey to the same illusions as people. Here, I invert the standard use of perceptual illusions to examine basic processing errors in current vision language models. I present these models with illusory-illusions, neighbors of common illusions that should not elicit processing errors. These include such things as perfectly reasonable ducks, crooked lines that truly are crooked, circles that seem to have different sizes because they are, in fact, of different sizes, and so on. I
arXiv:2405.12775v2 Announce Type: replace-cross Abstract: Discovering the semantics of multimodal utterances is essential for understanding human language and enhancing human-machine interactions. Existing methods manifest limitations in leveraging nonverbal information for discerning complex semantics in unsupervised scenarios. This paper introduces a novel unsupervised multimodal clustering method (UMC), making a pioneering contribution to this field. UMC introduces a unique approach to constructing augmentation views for multimodal data, which are then used to perform pre-training to establish well-initialized representations for subsequent clustering. An innovative strategy is proposed to dynamically select high-quality samples as guidance for representation learning, gauged by the density of each sample's nearest neighbors. Besides, it is equipped to automatically determine the optimal value for the top-$K$ parameter in each cluster to refine sample selection. Finally, both high-
arXiv:2607.13753v2 Announce Type: replace Abstract: Large language models have made strong reasoning gains through supervised fine-tuning, reinforcement learning, and on-policy distillation, yet these post-training methods are usually evaluated only by final-answer accuracy. We study how they reshape confidence during reasoning. We introduce a three-stage calibration framework that evaluates confidence before, during, and after chain-of-thought generation, corresponding to difficulty estimation, early termination, and answer aggregation. Through a controlled comparison on mathematical reasoning benchmarks, we find that OPD provides the most useful pre-reasoning confidence, SFT gives the strongest online signal for early stopping, and RL produces the most reliable trace-level signal for aggregation. We further show that confidence reliability is position-dependent: RL confidence becomes informative after a path-commitment phase, while OPD confidence is useful early but can become invers
arXiv:2607.00250v3 Announce Type: replace Abstract: Maltese OCR is constrained by the absence of a public, reusable paragraph-scale training corpus. We address this by generating synthetic Maltese line images, fine-tuning the Tesseract 5 LSTM, and combining five deterministic Tesseract configurations through anchor-preserving, lexicon-gated word-level arbitration. The method uses a fixed anchor stream, a longest-stream fallback, a confusion-based anchor corrector, and a Maltese-specific diacritic-restoration gate. Unlike canonical ROVER, candidate streams cannot restructure the anchor through insertions or deletions; they propose only eligible substitutions at aligned anchor positions. On the 422-paragraph development set of the DocEng 2026 Maltese OCR competition, the organizers' fine-tuned Tesseract baseline obtains CER 0.0234. Our pre-convention pipeline reaches CER 0.01317, a 44% reduction. Synthetic fine-tuning provides the largest single gain, while multi-stream arbitration contr
arXiv:2606.07559v2 Announce Type: replace Abstract: Fine-tuning a language model often fails silently when its correct completion must outrank a near-synonym competitor. Cross-entropy loss falls monotonically while the correct token never overtakes the competitor. We study five transformer architectures of two families over a fivefold parameter range, on ten contexts where the correct completion and nearest competitor overlap substantially. We construct an order parameter combining the predicted distribution with the geometric overlap between token embeddings. The density-matrix formulation is natural because the prediction distribution lives over a non-orthogonal embedding basis. It decomposes additively into a signal, tracking commitment to the correct token over its nearest competitor, and a background drag, set by how the embedding bulk leaks probability into the score. This isolates two failure modes. In kinematic failure the signal stays small and the model never commits. In stru
arXiv:2606.02780v3 Announce Type: replace Abstract: The success of the transformer architecture as the backbone of modern LLMs is in large part due to its use of attention layers. An attention layer follows the standard neural network paradigm: it takes the residual stream as input and thereby produces context-dependent query, key, and value vectors. However, we find that model performance meaningfully improves when deeper layers learn only a context-free value vector to preserve the original token information, without drawing on any context from the residual stream. When the model has access to this context-free value vector, adding back the context-dependent component provides little additional benefit for aggregate benchmark performance. Such context-free value vectors can be stored as sparse model parameters, eliminating the need to recompute or persistently cache these values. Through systematic ablations on the key design choices for such context-free value vectors, we propose Ba
arXiv:2605.29007v2 Announce Type: replace Abstract: Personalized tutoring, teacher preparation, and education research can benefit from worked errors annotated by the mechanisms that produced them. Authentic student errors with such cognitive labels are costly to collect and share, motivating the study of whether LLMs can generate taxonomy-targeted synthetic errors as complementary candidate material. We present a task-specific framework that generates errors targeted to a five-class Bloom-informed student-error taxonomy. A Generation Agent (GA) drafts a candidate erroneous solution conditioned on a target class, and an Examination Agent (EA) judges whether the draft is incorrect and class-consistent. The framework yields a reusable recipe for building class-stratified synthetic error datasets where authentic student corpora are unavailable. As a secondary diagnostic, targeted error generation is substantially harder than free-form incorrect-answer generation, and answer-grounding cont
arXiv:2605.14473v4 Announce Type: replace Abstract: Retrieval-Augmented Generation (RAG) is usually evaluated by whether the final answer is correct. Under knowledge conflict, this hides a key question: did the model follow retrieved evidence, rely on its parametric prior, or produce a post-hoc rationale? We study this as context compliance, the regime in which retrieved context controls the answer even when it conflicts with the model's prior knowledge. We introduce Context-Driven Decomposition (CDD), an inference-time diagnostic intervention that elicits contextual and prior answers, isolates the conflicting premise, and records a resolution trace that can be perturbed. Across Epi-Scale stress tests, TruthfulQA misconception injection, and cross-model reruns, CDD makes three behaviors visible. First, misleading retrieval can severely degrade accuracy: under a worst-case TruthfulQA misconception-injection probe, Standard RAG reaches only 15.0%. Second, better answers need not share th
arXiv:2605.12225v2 Announce Type: replace Abstract: While deep transformer-based models have advanced rapidly, their internal mechanisms remain largely a mystery. Recent work has prioritized understanding text-based transformer models, leaving ASR systems largely unexplored. In order to address this gap, we examine the internal representations of Whisper's encoder using a sparse autoencoder. We find diverse monosemantic features across linguistic and non-linguistic boundaries, spanning a hierarchy from phonetic to semantic representations, and conduct a causal feature-steering campaign across this hierarchy, including cross-lingual steering. We further find that steering is more reliable for higher-level features than lower-level ones, an asymmetry that may reflect redundant encoding of lower-level information. Altogether, this work demonstrates that Whisper's encoder represents a surprisingly rich hierarchy of linguistic information that extends well beyond what is strictly necessary
arXiv:2605.01224v2 Announce Type: replace Abstract: This paper argues that contemporary multilingual NLP has converged on a fragile and misleading paradigm of incidental multilingualism. Today's LLMs appear multilingual largely because they are trained on massive, uneven web corpora, not because multilingual or multicultural competence has been treated as a core design objective. We contend that this paradigm systematically produces unequal, brittle, and opaque behavior across languages, with severe consequences in real-world and agentic deployments where models must reason, plan, and act across multiple linguistic contexts. We report a focused empirical study of two practical questions: which languages models self-report as supported and which languages they actually respond in across multilingual prompts. We additionally demonstrate how even a simple language-change attack can surface these failures and expose hidden assumptions about language in LLM-based systems. To address this, w
arXiv:2604.09497v2 Announce Type: replace Abstract: Accurate evaluation is central to the large language model (LLM) ecosystem, guiding model selection and downstream adoption across diverse use cases. In practice, however, evaluating generative outputs typically relies on rigid lexical methods to extract and assess answers, which can conflate a model's true problem-solving ability with its compliance with predefined formatting guidelines. While recent LLM-as-a-Judge approaches mitigate this issue by assessing semantic correctness rather than strict structural conformity, they also introduce substantial computational overhead, making evaluation costly. In this work, we first systematically investigate the limitations of lexical evaluation through a large-scale empirical study spanning 36 models and 15 downstream tasks, demonstrating that such methods correlate poorly with human judgments. To address this limitation, we introduce BERT-as-a-Judge, an encoder-driven approach for assessing
arXiv:2604.08879v2 Announce Type: replace Abstract: Moving beyond the traditional binary classification paradigm of Multimodal Sarcasm Detection, Multimodal Sarcasm Target Identification (MSTI) presents a more formidable challenge, requiring precise localization of fine-grained targets such as textual phrases and visual regions. Existing approaches predominantly rely on implicit cross-modal alignment, offering limited interpretability and suboptimal fine-grained localization. To address these limitations, we propose GRASP, Grounded Chain-of-Thought ReAsoning with Dual-Stage Optimization for Multimodal Sarcasm Prediction and Target Identification, a framework that integrates visual grounding with explicit Chain-of-Thought (CoT) reasoning to move beyond black-box MSTI. Specifically, we curate MSTI-MAX, a refined dataset that mitigates class imbalance and enriches multimodal sarcasm cues. We introduce Grounded CoT reasoning, which explicitly anchors sarcasm-related visual regions within t
arXiv:2604.08523v2 Announce Type: replace Abstract: AI agents may be able to assist with emails and documents, but can they reliably complete everyday online workflows on real websites? Everyday online tasks offer a realistic yet unsolved testbed for evaluating the next generation of AI agents. To this end, we introduce ClawBench, an evaluation framework comprising 153 everyday online tasks that people need to accomplish regularly in their lives and work, spanning 144 platforms across 15 categories, from completing purchases and booking appointments to submitting job applications. These tasks require capabilities beyond existing benchmarks, such as obtaining relevant information from user-provided documents, navigating multi-step workflows across diverse platforms, and write-heavy operations like filling in many detailed forms correctly. Unlike existing benchmarks that evaluate agents in offline sandboxes with static pages, ClawBench operates on production websites, preserving the full
arXiv:2604.01457v2 Announce Type: replace Abstract: Large language models are often not just wrong, but \emph{confidently wrong}: when they produce factually incorrect answers, they tend to verbalize overly high confidence rather than signal uncertainty. Such verbalized overconfidence can mislead users and weaken confidence scores as a reliable uncertainty signal, yet its internal mechanisms remain poorly understood. We present a circuit-level mechanistic analysis of this inflated verbalized confidence in LLMs, organized around three axes: capturing verbalized confidence as a differentiable internal signal, identifying the circuits that causally inflate it, and leveraging these insights for targeted inference-time recalibration. Across two instruction-tuned LLMs on three datasets, we find that a compact set of MLP blocks and attention heads, concentrated in middle-to-late layers, consistently writes the confidence-inflation signal at the final token position. We further show that targe
arXiv:2603.27752v2 Announce Type: replace Abstract: Large language models can still hallucinate in retrieval-augmented generation (RAG), producing claims that are unsupported by or conflict with the retrieved context. Detecting such errors remains challenging when faithfulness is judged solely against the retrieved context: many existing detectors return holistic answer-level scores, while others target open-domain factuality or fail to provide evidence-grounded diagnostics. We present RT4CHART, a retromorphic testing framework for context-faithfulness assessment. RT4CHART decomposes an answer into independently verifiable claims, performs hierarchical local-to-global verification against the retrieved context, and assigns each claim one of three labels: entailed, contradicted, or baseless. It further maps these claim-level decisions back to specific answer spans and returns explicit context-side evidence, enabling fine-grained auditing rather than opaque scoring. We evaluate RT4CHART
arXiv:2603.20843v3 Announce Type: replace Abstract: Long-context language modeling is commonly framed as a scalability challenge of token-level attention, yet local-to-global information structuring remains largely implicit in existing approaches. Drawing on cognitive theories of discourse comprehension, we propose HiCI (Hierarchical Construction--Integration), a hierarchical attention module that constructs segment-level representations, integrates them into a shared global context, and broadcasts both to condition segment-level attention. We validate HiCI through parameter-efficient adaptation of LLaMA-2 with only <5.5% additional parameters, extending context from 4K to 100K tokens (7B) and 64K tokens (13B). Across language modeling, retrieval, and instruction-following benchmarks, HiCI yields consistent improvements over strong baselines, including matching proprietary models on topic retrieval and surpassing GPT-3.5-Turbo-16K on code comprehension. These results demonstrate the ef
arXiv:2603.14355v2 Announce Type: replace Abstract: Safety tuning through supervised fine-tuning and reinforcement learning from human feedback has substantially improved the robustness of large language models. However, it typically suppresses rather eliminates unsafe behaviors, leaving rare but critical failures hidden in the long tail of the output distribution. While most red-teaming work emphasizes adversarial prompt search, we show that these hidden risks can be systematically exposed through diverse response generation. Specifically, we show that, for a fixed safety-critical prompt, increasing the number and diversity of sampled responses monotonically raises the jailbreak success rate. To efficiently uncover these failures, we propose Progressive Diverse Population Sampling (PDPS). This approach replaces naive, large-scale IID sampling with a multi-stage expansion-and-selection strategy that generates a compact, semantically diverse set of responses at a substantially lower com
arXiv:2602.11961v3 Announce Type: replace Abstract: Open large language models (LLMs) have demonstrated improving multilingual capabilities in recent years. In this paper, we present a study of open LLMs for multilingual machine translation (MT) across a range of languages, and investigate the effects of model scaling and data scaling when adapting open LLMs to multilingual MT through continual pretraining and instruction finetuning. Based on the Gemma3 model family, we develop MiLMMT-46, which achieves top-tier multilingual translation performance across 46 languages. Extensive experiments show that MiLMMT-46 consistently outperforms recent state-of-the-art (SOTA) models, including Seed-X, HY-MT-1.5, and TranslateGemma, and achieves competitive performance with strong proprietary systems such as Google Translate and Gemini 3 Pro. Models are released at https://huggingface.co/collections/xiaomi-research/milmmt-46. Codes are released at https://github.com/xiaomi-research/gemmax.
arXiv:2602.10382v3 Announce Type: replace Abstract: Backdoor attacks pose significant security risks for Large Language Models (LLMs), yet the internal mechanisms by which triggers operate remain poorly understood. We present the first mechanistic analysis of trigger-induced language-switching backdoors injected during pre-training, studying the Gaperon model family (1B, 8B and 24B). Using activation patching, we localize trigger formation and identify which attention heads process trigger and natural language information. Our central finding is that trigger heads substantially overlap with heads naturally encoding output language across model scales, with Jaccard indices between 0.18 and 0.43 over the top 10 heads identified. This suggests that backdoor triggers do not form new circuits but instead co-opt the model's existing language components and representations. These findings have implications for backdoor defense as detection methods and mitigation strategies could leverage this
arXiv:2601.13359v3 Announce Type: replace Abstract: Prefill attacks are an effective and low-cost jailbreaking method, as they directly insert an acceptance sequence (e.g., "Sure, here is...") at the start of an LLM's output and lead the model to continue the response. We make two contributions to this prior work. First, we show that an unsophisticated adversary can improve the well-known prefill attacks by ensembling a small number of prefill variants. Running three easy-to-generate prefills yields a combined attack success rate (ASR) of 22%, 90%, and 99% on Gemma-7B, Llama-3.1-8B, and Qwen3-8B respectively, an up to 38 percentage point improvement over the standard "Sure, here's..." prefill and up to 82 percentage points over our reproduction of GCG (Zou et al., 2023). Second, we introduce "sockpuppetting", a hybrid attack that optimizes an adversarial suffix placed inside the "assistant" message block of the chat template, rather than within the user prompt. The rolling variant of t
arXiv:2601.07121v2 Announce Type: replace Abstract: Large language models (LLMs) are increasingly used not only for problem solving but also for creative ideation; however, generating ideas that are both novel and coherent remains challenging. While high-temperature sampling can promote originality, it often compromises consistency and usefulness. Here, we propose ReMIND, a four-stage framework comprising wake, which establishes a stable semantic baseline through low-temperature generation; dream, which performs high-temperature exploratory generation; judge, which evaluates candidate outputs for consistency and extracts salient novel ideas; and rewake, which consolidates selected ideas into coherent final outputs. By assigning these functions to independent LLM modules, ReMIND explicitly separates exploration from stabilization. We systematically evaluated diverse model configurations using multiple creative ideation tasks. External evaluations showed that novelty enhancement could em
arXiv:2601.03481v2 Announce Type: replace Abstract: Existing hate speech detection models are often opaque and rely on surface-level lexical cues, which makes them vulnerable to spurious correlations and limits robustness, interpretability and cultural contextualization. We propose Supervised Moral Rationale Attention (SMRA), the first self-explaining hate speech detection framework to incorporate moral rationales as direct supervision for attention alignment. Based on Moral Foundations Theory, SMRA aligns token-level attention with expert-annotated moral rationales, guiding models to attend to morally salient spans. Unlike prior rationale-supervised or post-hoc approaches, SMRA integrates moral rationale supervision directly into the training objective, producing inherently interpretable and contextualized explanations. To support our framework, we also introduce HateBRMoralXplain, a Brazilian Portuguese benchmark dataset annotated with hate labels, moral categories, token-level moral
arXiv:2601.03043v4 Announce Type: replace Abstract: Large language models (LLMs) demonstrate strong capabilities across a wide range of complex tasks and are increasingly deployed at scale, placing significant demands on inference efficiency. Prior work typically decomposes inference into prefill and decode stages, with the decode stage dominating total latency. To reduce time and memory complexity in the decode stage, a line of work introduces sparse-attention algorithms. In this paper, we show, both empirically and theoretically, that sparse attention can paradoxically increase end-to-end complexity: information loss often induces significantly longer sequences, a phenomenon we term ``Less is Less'' (Lil). To mitigate the Lil problem, we propose an early-stopping algorithm that detects the threshold where information loss exceeds information gain during sparse decoding. Our early-stopping algorithm reduces token consumption by up to 90% with a marginal accuracy degradation of less th
arXiv:2512.02924v3 Announce Type: replace Abstract: While Neural Processing Units (NPUs) offer high theoretical efficiency for edge AI, state-of-the-art Vision--Language Models (VLMs) tailored for GPUs often falter on these substrates. We attribute this hardware-model mismatch to two primary factors: the quantization brittleness of Vision Transformers (ViTs) and the I/O-bound nature of autoregressive attention mechanisms, which fail to utilize the high arithmetic throughput of NPUs. To bridge this gap, we propose AutoNeural, an NPU-native VLM architecture co-designed for integer-only inference. We replace the standard ViT encoder with a MobileNetV5-style backbone utilizing depthwise separable convolutions, which ensures bounded activation distributions for stable INT4/8/16 quantization. Complementing this, our language backbone integrates State-Space Model (SSM) principles with Transformer layers, employing efficient gated convolutions to achieve linear-time complexity. This hybrid des