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:2605.11836v2 Announce Type: replace-cross Abstract: Lifelong Model Editing aims to continuously update evolving facts in Large Language Models while preserving unrelated knowledge and general capabilities, yet it remains plagued by catastrophic forgetting and model collapse. Empirically, we find that recent editors resilient over long horizons share the same core strategy: Lifelong Normalization (LN), which normalizes value gradients using running statistics. Removing LN causes immediate performance collapse, and we observe a counter-intuitive positive cumulative effect where early edits can promote the success of future edits. Yet the mechanism of LN remains a "black box", leaving its precise role in lifelong stability poorly understood. In this work, we provide the first theoretical account of LN in the lifelong regime. Our analysis reveals a self-reinforcing stability loop and proves that, when combined with ridge-regularized regression, LN yields parameter updates with asympt
arXiv:2605.05482v2 Announce Type: replace-cross Abstract: Large language models (LLMs) are rapidly being adopted across various domains. However, their adoption in banking industry faces resistance due to demands for high accuracy, regulatory compliance, and the need for verifiable and grounded responses. We present a unified, data-efficient framework for training grounded domain-specific LLMs that optimizes answer quality, citation grounding, and calibrated refusal under real-world deployment constraints. First, we describe a data generation pipeline that combines LLM-as-a-Judge filtering, citation annotation, and curriculum learning with only 143M tokens. The resulting 12B model achieves high answer quality outperforming GPT-4.1 on citation grounding, with a modest citation tradeoff versus the untuned base. Second, we propose a calibrated refusal mechanism: training on 22% unanswerable examples yield a 12% "I don't know" rate, substantially improving over the base model's unsafe 4.3%
arXiv:2604.08571v3 Announce Type: replace-cross Abstract: While Large Language Models (LLMs) achieve high performance on standard mathematical benchmarks, their problem-solving abilities depend on the context and textual formatting. We introduce the Robust Reasoning Benchmark (RRB), a pipeline of 13 deterministic textual perturbations applied to AIME 2024 and AIME 2025. Evaluating 8 state-of-the-art models, we find that frontier models are largely resilient, with the notable exception of Claude, which categorically refuses many transformed prompts. Open-weights reasoning models exhibit a range of failure modes under structural noise (cognitive thrashing, tokenization breakdown, and reasoning collapse), with up to 54% average accuracy drops across perturbations and up to 100% on some. We further study one of these failure modes in isolation: attention dilution caused by the model's own chain-of-thought. By tasking models with solving multiple independent mathematical problems sequential
arXiv:2603.21014v2 Announce Type: replace-cross Abstract: Mechanistic interpretability seeks to understand how Large Language Models (LLMs) represent and process information. Recent approaches based on dictionary learning and transcoders enable representing model computation in terms of sparse, interpretable features and their interactions, giving rise to feature attribution graphs. However, these graphs are often large and redundant, limiting their interpretability in practice. Cross-Layer Transcoders (CLTs) address this issue by sharing features across layers while preserving layer-specific decoding, yielding more compact representations, but remain difficult to train and analyze at scale. We introduce an open-source library for end-to-end training and interpretability of CLTs. Our framework integrates scalable distributed training with model sharding and compressed activation caching, a unified automated interpretability pipeline for feature analysis and explanation, attribution gra
arXiv:2602.22647v2 Announce Type: replace-cross Abstract: Generative retrieval has emerged as a powerful paradigm for LLM-based recommendation. However, industrial recommender systems often benefit from restricting the output space to a constrained subset of items based on business logic (e.g. enforcing content freshness or product category), which standard autoregressive decoding cannot natively support. Moreover, existing constrained decoding methods that make use of prefix trees (Tries) incur severe latency penalties on hardware accelerators (TPUs/GPUs). In this work, we introduce STATIC (Sparse Transition Matrix-Accelerated Trie Index for Constrained Decoding), an efficient and scalable constrained decoding technique designed specifically for high-throughput LLM-based generative retrieval on TPUs/GPUs. By flattening the prefix tree into a static Compressed Sparse Row (CSR) matrix, we transform irregular tree traversals into fully vectorized sparse matrix operations, unlocking massi
arXiv:2607.14480v2 Announce Type: replace Abstract: LLM evaluators (trained reward models and prompted LLM-as-a-Judge) are routinely validated via pairwise accuracy. In a multilingual setting, this operates under the premise that high pairwise accuracy implies reliable, language-neutral scoring. We show that this assumption does not hold. We conduct experiments with semantically identical instruction-response pairs across 23 languages, and find that multilingual evaluators assign significantly different scores to different evaluation languages. The bias is statistically significant and consistent across eight open-weight evaluators of different architectures and training paradigms, persists in frontier judges, and is strongly correlated with language resource level: lower-resource languages are scored more generously. Meanwhile, these biases are invisible to pairwise accuracy: evaluators achieve above 90% pairwise accuracy, yet have up to 43% difference in acceptance rate across langua
arXiv:2606.11018v2 Announce Type: replace Abstract: Measuring subjective constructs in naturally occurring social media text requires annotation procedures that are theoretically grounded, empirically validated, and transferable to an encoder model for scalable prediction. Using non-English social media posts annotated according to Schwartz's theory of basic human values, we investigate how different LLMs, prompts, and instruction languages operationalize the expression of values in text. We argue that although texts may permit multiple plausible interpretations, theory-based value definitions can constrain interpretations and reduce spurious value attributions. Beyond precision, recall, and F1, we evaluate structural alignment between values, error structure, confidence-ambiguity relations, and annotation stability. We show that different LLMs produce different value interpretations. Iterative prompt calibration through error analysis reduces misattributions and improves alignment wit
arXiv:2604.27039v2 Announce Type: replace Abstract: Tokens are the fundamental units of computation in modern autoregressive models, and generation length directly influences both inference cost and reasoning performance. Despite its importance, existing approaches model length primarily at the coarse sequence level. We introduce the Length Value Model (LenVM), a token-level framework that estimates the remaining generation length at every decoding step. By formulating length modeling as a value estimation problem and assigning a constant negative reward to each generated token, LenVM predicts a bounded, discounted return that is a monotone proxy for the remaining generation horizon. This value formulation provides annotation-free, dense, unbiased, and scalable supervision. Experiments on LLMs and VLMs show that LenVM supports exact control, continuous performance--efficiency steering, length prediction, and interpretation. On LIFEBench-token, it raises the exact-length score of Qwen2.
arXiv:2604.24927v2 Announce Type: replace Abstract: Generating diverse responses is crucial for test-time scaling of large language models (LLMs), yet standard stochastic sampling mostly yields surface-level lexical variation, limiting semantic exploration. In this paper, we propose Exploratory Sampling (ESamp), a decoding approach that explicitly encourages semantic diversity during generation. ESamp is motivated by the well-known observation that neural networks tend to make lower-error predictions on inputs similar to those encountered before, and incur higher prediction error on novel ones. Building on this property, we train a lightweight Distiller at test time to predict deep-layer hidden representations of the LLM from its shallow-layer representations to model the LLM's depth-wise representation transitions. During decoding, the Distiller continuously adapts to the mappings induced by the current generation context. ESamp uses the prediction error as a novelty signal to reweigh
arXiv:2603.25821v2 Announce Type: replace Abstract: We present Doctorina MedBench, a comprehensive evaluation framework for agent-based medical AI based on the simulation of realistic physician-patient interactions. Unlike traditional medical benchmarks that rely on solving standardized test questions, the proposed approach models a multi-step clinical dialogue in which either a physician or an AI system must collect medical history, analyze attached materials (including laboratory reports, images, and medical documents), formulate differential diagnoses, and provide personalized recommendations. System performance is evaluated using the D.O.T.S. metric, which consists of four components: Diagnosis, Observations/Investigations, Treatment, and Step Count, enabling assessment of both clinical correctness and dialogue efficiency. The system also incorporates a multi-level testing and quality monitoring architecture designed to detect model degradation during both development and deploymen
arXiv:2603.24917v2 Announce Type: replace Abstract: Recent work shows that standard greedy-decoding extraction methods for quantifying memorization in LLMs miss how extraction risk varies across sequences. Probabilistic extraction -- computing the probability of generating a target suffix given a prefix under a decoding scheme -- addresses this, but is tractable only for verbatim memorization, missing near-verbatim instances that pose similar privacy and copyright risks. Quantifying near-verbatim extraction risk is expensive: the set of near-verbatim suffixes is combinatorially large, and reliable Monte Carlo (MC) estimation can require ~100,000 samples per sequence. To mitigate this cost, we introduce decoding-constrained beam search, which yields deterministic lower bounds on near-verbatim extraction risk at a cost comparable to ~20 MC samples per sequence. Across experiments, our approach surfaces information invisible to verbatim methods: many more extractable sequences, substantia
arXiv:2603.16410v2 Announce Type: replace Abstract: Creative plot generation presents a fundamental challenge for language models: transforming a concise premise into a coherent narrative that sustains global coherence, character development, pacing, tone consistency, and emotional progression. Although recent Large Language Models (LLMs) demonstrate strong fluency on general-purpose tasks, they require preference alignment to perform well on domain-specific tasks such as creative plot generation. However, conducting such alignment at the scale of frontier LLMs is computationally prohibitive, significantly limiting accessibility and practical deployment. To address this, we present PlotTwist, a structured framework that enables Small Language Models (SLMs) with $\leq$3B active parameters to generate high-quality, premise-conditioned plots competitive with frontier systems of vastly greater parameter scale. Our approach decomposes generation into three specialized components: (1) an Asp
arXiv:2603.16299v2 Announce Type: replace Abstract: We introduce PyPhonPlan, a Python toolkit for implementing dynamical models of phonetic planning using coupled dynamic neural fields and task dynamic simulations. The toolkit provides modular components for defining planning, perception and memory fields, as well as between-field coupling, gestural inputs, and using field activation profiles to solve tract variable trajectories. We illustrate the toolkit's capabilities through an example application: simulating production/perception loops with a coupled memory field, which demonstrates the framework's ability to model interactive speech dynamics using representations that are temporally-principled, neurally-grounded, and phonetically-rich. PyPhonPlan is released as open-source software and contains executable examples to promote reproducibility, extensibility, and cumulative computational development for speech communication research.
arXiv:2602.05493v2 Announce Type: replace Abstract: Data annotation remains a significant bottleneck in the field of humanities and social sciences, particularly for complex linguistic tasks such as metaphor identification. While Large Language Models (LLMs) show promise, a significant gap remains between the theoretical capability of LLMs and their practical utility for researchers. This paper introduces LinguistAgent, an integrated, user-friendly platform that leverages a reflective multi-model architecture to automate linguistic annotation. The platform comprises an Annotator and an optional Reviewer to simulate a peer-review process. This platform supports comparative experiments across three main paradigms: Prompt Engineering (Zero-shot/Few-shot/Chain-of-thought), Retrieval-Augmented Generation, and Fine-tuning. We demonstrate LinguistAgent's efficacy by replicating the task of metaphor identification from a published study, which provides real-time token-level evaluation (F1 and
arXiv:2601.13024v2 Announce Type: replace Abstract: Culture serves as a fundamental determinant of human affective processing and profoundly shapes how individuals perceive and interpret emotional stimuli. Despite this intrinsic link extant evaluations regarding cultural alignment within Large Language Models primarily prioritize declarative knowledge such as geographical facts or established societal customs. These benchmarks remain insufficient to capture the subjective interpretative variance inherent to diverse sociocultural lenses. To address this limitation, we introduce CEDAR, a multimodal benchmark constructed entirely from scenarios capturing Culturally \underline{\textsc{E}}licited \underline{\textsc{D}}istinct \underline{\textsc{A}}ffective \underline{\textsc{R}}esponses. To construct CEDAR, we implement a novel pipeline that leverages LLM-generated provisional labels to isolate instances yielding cross-cultural emotional distinctions, and subsequently derives reliable groun
arXiv:2601.06471v2 Announce Type: replace Abstract: Large language model (LLM) personalization aims to adapt general-purpose models to individual users. Most existing methods, however, are developed under data-rich and resource-abundant settings, often incurring privacy risks. In contrast, realistic personalization typically occurs after deployment under (i) extremely limited user data, (ii) constrained computational resources, and (iii) strict privacy requirements. We propose PRISP, a lightweight and privacy-safe personalization framework tailored to these constraints. PRISP leverages a Text-to-LoRA hypernetwork to generate task-aware LoRA parameters from task descriptions, and enables efficient user personalization by optimizing a small subset of task-aware LoRA parameters together with minimal additional modules using few-shot user data. Experiments on a few-shot variant of the LaMP benchmark demonstrate that PRISP achieves strong overall performance compared to prior approaches, wh
arXiv:2511.19166v4 Announce Type: replace Abstract: Large language models (LLMs) are widely used as information sources, yet small changes in semantic assumptions can destabilize their beliefs. We introduce P-StaT (Perturbation Stability of Truth), a framework for evaluating belief stability under matched semantic perturbations in both representational and behavioral settings. Across 21 LLMs and three domains, we compare perturbations involving familiar Fictional statements against synthetically generated and unfamiliar Synthetic statements. Unfamiliar Synthetic perturbations generally induce greater epistemic instability than familiar Fictional perturbations, with behavioral belief retraction rates frequently exceeding 0.5 (50%). Finally, exploratory clustering analyses reveal recurring themes among retracted statements, including ambiguity, technical terminology, and obscure concepts. These results show that epistemic familiarity is systematically associated with stability under sema
arXiv:2510.02345v4 Announce Type: replace Abstract: Mixture-of-Experts (MoE) Large Language Models (LLMs) face a trilemma of load imbalance, parameter redundancy, and communication overhead. We introduce a unified framework based on dynamic expert clustering and structured compression to address these issues cohesively. Our method employs an online clustering procedure that periodically regroups experts using a fused metric of parameter and activation similarity, which stabilizes expert utilization. To our knowledge, this is one of the first frameworks to leverage the semantic embedding capability of the router to dynamically reconfigure the model's architecture during training for substantial efficiency gains. Within each cluster, we decompose expert weights into a shared base matrix and extremely low-rank residual adapters, achieving up to fivefold parameter reduction per group while preserving specialization. This structure enables a two-stage hierarchical routing strategy: tokens a
arXiv:2509.25045v3 Announce Type: replace Abstract: Despite their capabilities, Large Language Models (LLMs) remain opaque with limited understanding of their internal representations. Current interpretability methods either focus on input-oriented feature extraction, such as supervised probes and Sparse Autoencoders (SAEs), or on output distribution inspection, such as logit-oriented approaches. A full understanding of LLM vector spaces, however, requires integrating both perspectives, something existing approaches struggle with due to constraints on latent feature definitions. We introduce the Hyperdimensional Probe, a hybrid supervised probe that combines symbolic representations with neural probing. Leveraging Vector Symbolic Architectures (VSAs) and hypervector algebra, it unifies prior methods: the top-down interpretability of supervised probes, SAE's sparsity-driven proxy space, and output-oriented logit investigation. By combining the supervised learning paradigm of traditional
arXiv:2509.23381v2 Announce Type: replace Abstract: We introduce Guard Vector, a safety task vector computed as the parameter difference between a guardrail model (Guard Model) and a same-architecture pretrained language model. Composing this vector with a target language model yields a Target Guard Model (TGM). We then adapt TGM with a streaming-aware approach that combines prefix-based training and evaluation with a classifier that produces a single-token output. With this composition alone, TGM improves classification quality over established Guard Models across standard safety suites and enables language extensibility to Chinese, Japanese, and Korean, requiring neither additional training nor target language labels for this composition step. It also demonstrates model portability across two widely used public guardrail backbones, Llama and Gemma. With prefix SFT (supervised fine-tuning), TGM preserves classification quality under streaming by aligning the behavior between prefix in
arXiv:2508.06374v3 Announce Type: replace Abstract: With the surge of large language models (LLMs) and their ability to produce customized output, style-personalized text generation--"write like me"--has become a rapidly growing area of interest. However, style personalization is highly specific, relative to every user, and depends strongly on the pragmatic context, which makes it uniquely challenging. Although prior research has introduced benchmarks and metrics for this area, they tend to be non-standardized and have known limitations (e.g., poor correlation with human subjects). LLMs have been found to not capture author-specific style well, it follows that the metrics themselves must be scrutinized carefully. In this work we critically examine the effectiveness of the most common metrics used in the field, such as BLEU, embeddings, and LLMs-as-judges. We evaluate these metrics using our proposed style discrimination benchmark, which spans eight diverse writing tasks across three ev
arXiv:2508.00429v5 Announce Type: replace Abstract: Graph Neural Networks (GNNs) have achieved remarkable success in graph-based learning by propagating information among neighbor nodes via predefined aggregation mechanisms. However, such fixed schemes often suffer from two key limitations. First, they cannot handle the imbalance in node informativeness -- some nodes are rich in information, while others remain sparse. Second, predefined message passing primarily leverages local structural similarity while ignoring global semantic relationships across the graph, limiting the model's ability to capture distant but relevant information. We propose Retrieval-augmented Graph Agentic Network (ReaGAN), an agent-based framework that empowers each node with autonomous, node-level decision-making. Each node acts as an agent that independently plans its next action based on its internal memory, enabling node-level planning and adaptive message propagation. Additionally, retrieval-augmented gener
arXiv:2506.18421v3 Announce Type: replace Abstract: The majority of data in businesses and industries is stored in tables, databases, and data warehouses. Reasoning with table-structured data poses significant challenges for large language models (LLMs) due to its hidden semantics, inherent complexity, and structured nature. One of these challenges is lacking an effective evaluation benchmark fairly reflecting the performances of LLMs on broad table reasoning abilities. In this paper, we fill in this gap by presenting a comprehensive table reasoning benchmark, TReB. Firstly, we propose a taxonomy to systematically measure both shallow table understanding abilities and deep table reasoning abilities, covering a total of 26 sub-tasks. We then construct a high quality dataset through a dedicated data processing and synthesis procedure. Based on these well-constructed samples, we design an evaluation framework to robustly measure table reasoning capabilities with three distinct inference m
arXiv:2505.24302v3 Announce Type: replace Abstract: Large Language Models (LLMs) are increasingly used to support scientific research, but their knowledge of scientific advancements can quickly become outdated. We introduce ScienceMeter, a new framework for evaluating scientific knowledge update methods over scientific knowledge spanning the past, present, and future. ScienceMeter defines three metrics: knowledge preservation, the extent to which models' understanding of previously learned papers is preserved; knowledge acquisition, how well scientific claims from newly introduced papers are acquired; and knowledge projection, the ability of the updated model to anticipate or generalize to related scientific claims that may emerge in the future. Using ScienceMeter, we evaluate the scientific knowledge of LLMs through claim judgment and generation tasks on a curated dataset across ten domains. We evaluate five representative knowledge update approaches and find that the best-performing
arXiv:2502.19737v2 Announce Type: replace Abstract: We introduce XCOMPS in this work, a multilingual conceptual minimal pair dataset covering 17 languages. Using this dataset, we evaluate LLMs' multilingual conceptual understanding through metalinguistic prompting, direct probability measurement, and neurolinguistic probing. By comparing base, instruction-tuned, and knowledge-distilled models, we find that: 1) LLMs exhibit weaker conceptual understanding for low-resource languages, and accuracy varies across languages despite being tested on the same concept sets. 2) LLMs excel at distinguishing concept-property pairs that are visibly different but exhibit a marked performance drop when negative pairs share subtle semantic similarities. 3) Instruction tuning improves performance in concept understanding but does not enhance internal competence; knowledge distillation can enhance internal competence in conceptual understanding for low-resource languages with limited gains in explicit ta
arXiv:2409.07314v3 Announce Type: replace Abstract: While Large Language Models (LLMs) achieve superhuman performance on standardized medical licensing exams, these static benchmarks have become saturated and increasingly disconnected from the functional requirements of clinical workflows. To bridge the gap between theoretical capability and verified utility, we introduce MEDIC, a comprehensive evaluation framework establishing leading indicators of clinical LLM competence across five dimensions. These upfront indicators reveal cross-benchmark capability gaps, such as the divergence between static knowledge retrieval and functional execution, that inform model selection before costly deployment-based evaluation. Beyond standard question-answering, we assess operational capabilities using deterministic execution protocols and a novel Cross-Examination Framework (CEF), which quantifies information fidelity and hallucination rates without reliance on reference texts. Our evaluation across
arXiv:2402.18121v2 Announce Type: replace Abstract: This study assesses four cutting-edge language models in the underexplored Aminoacian language. Through evaluation, it scrutinizes their adaptability, effectiveness, and limitations in text generation, semantic coherence, and contextual understanding. Uncovering insights into these models' performance in a low-resourced language, this research pioneers pathways to bridge linguistic gaps. By offering benchmarks and understanding challenges, it lays groundwork for future advancements in natural language processing, aiming to elevate the applicability of language models in similar linguistic landscapes, marking a significant step toward inclusivity and progress in language technology.
arXiv:2106.16038v2 Announce Type: replace Abstract: Recent pretraining models in Chinese neglect two important aspects specific to the Chinese language: glyph and pinyin, which carry significant syntax and semantic information for language understanding. In this work, we propose ChineseBERT, which incorporates both the {\it glyph} and {\it pinyin} information of Chinese characters into language model pretraining. The glyph embedding is obtained based on different fonts of a Chinese character, being able to capture character semantics from the visual features, and the pinyin embedding characterizes the pronunciation of Chinese characters, which handles the highly prevalent heteronym phenomenon in Chinese (the same character has different pronunciations with different meanings). Pretrained on large-scale unlabeled Chinese corpus, the proposed ChineseBERT model yields significant performance boost over baseline models with fewer training steps. The porpsoed model achieves new SOTA perform
arXiv:1911.02855v4 Announce Type: replace Abstract: Many NLP tasks such as tagging and machine reading comprehension are faced with the severe data imbalance issue: negative examples significantly outnumber positive examples, and the huge number of background examples (or easy-negative examples) overwhelms the training. The most commonly used cross entropy (CE) criteria is actually an accuracy-oriented objective, and thus creates a discrepancy between training and test: at training time, each training instance contributes equally to the objective function, while at test time F1 score concerns more about positive examples. In this paper, we propose to use dice loss in replacement of the standard cross-entropy objective for data-imbalanced NLP tasks. Dice loss is based on the Sorensen-Dice coefficient or Tversky index, which attaches similar importance to false positives and false negatives, and is more immune to the data-imbalance issue. To further alleviate the dominating influence fro
arXiv:1905.05529v5 Announce Type: replace Abstract: In this paper, we propose a new paradigm for the task of entity-relation extraction. We cast the task as a multi-turn question answering problem, i.e., the extraction of entities and relations is transformed to the task of identifying answer spans from the context. This multi-turn QA formalization comes with several key advantages: firstly, the question query encodes important information for the entity/relation class we want to identify; secondly, QA provides a natural way of jointly modeling entity and relation; and thirdly, it allows us to exploit the well developed machine reading comprehension (MRC) models. Experiments on the ACE and the CoNLL04 corpora demonstrate that the proposed paradigm significantly outperforms previous best models. We are able to obtain the state-of-the-art results on all of the ACE04, ACE05 and CoNLL04 datasets, increasing the SOTA results on the three datasets to 49.4 (+1.0), 60.2 (+0.6) and 68.9 (+2.1),
arXiv:2607.19336v1 Announce Type: cross Abstract: Agentic systems large language model (LLM) based architectures capable of reasoning, planning, acting, and coordinating with tools and other agents are rapidly transitioning from research prototypes to production scale deployments across domains such as software engineering, scientific discovery, and finance. While academic work has emphasized benchmarks and algorithmic innovation, deployment raises new challenges around robustness, safety, and reliability. This tutorial brings together researchers and practitioners to explore advances in reasoning and planning, multi agent coordination, and evaluation, highlighting open challenges arising from deployment experience. Through applied case studies in pharmaceutical discovery and financial systems, we analyze common design patterns that make agentic systems successful, and discuss practical mitigation strategies for failure modes, such as verification pipelines, fallback mechanisms, and hu
arXiv:2607.19317v1 Announce Type: cross Abstract: Circuit analysis can support not only model explanation but also downstream interventions such as pruning, editing, steering, and selective fine-tuning. However, conducting such analyses currently requires stitching together separate implementations for discovery, evaluation, and intervention, as well as hand-authoring the contrastive prompts required by many discovery methods. This fragmentation makes methods difficult to compare and limits their application beyond canonical tasks. We introduce CircuitKIT, a source-available library that connects the circuit-analysis workflow through a typed, serializable representation. CircuitKIT provides a suite of discovery algorithms, declarative interfaces for mapping structured data into discovery tasks, complementary circuit diagnostics, and downstream application modules. Together, these components provide common infrastructure for conducting and comparing circuit analyses. The library, exampl
arXiv:2607.19223v1 Announce Type: cross Abstract: Speculative decoding, in which a lightweight draft model first generates a draft sequence that is then verified in parallel by the target model, has become a prevalent paradigm for accelerating large language model inference. Recent work such as DFlash further boosts drafting efficiency by leveraging diffusion drafters, whose parallel denoising mechanism enables draft generation in a single forward pass. In this work, we uncover a central pitfall of diffusion drafters: bidirectional attention is a double-edged sword. On one hand, it endows the model with parallel generation and global contextual modeling capabilities; on the other hand, this inherent global dependency introduces high variance at both the domain-level and the token-level: acceptance rates fluctuate substantially across different domains, and draft token quality also varies heterogeneously at different token positions. To tackle this issue, we propose AdaFlash framework,
arXiv:2607.19096v1 Announce Type: cross Abstract: Agent-memory workloads mix direct factual lookup, relation-chain and current-state reasoning, and broad synthesis over long histories. We describe Supra Cognitive Modes (SCM), an architecture that maps explicit or automatically selected per-query modes to retrieval and synthesis payloads over one shared ingest substrate. A frozen semantic classifier and runtime gates dispatch queries among fused lexical and dense lookup, graph or iterative multi-hop handling, and stratified long-form synthesis. The substrate combines multi-granularity embeddings, extracted triples, fact-version metadata, and optional asynchronous enrichments. We characterize the deployed configuration on three benchmarks: Long-term Conversational Memory (LoCoMo; n = 1,986), MemoryAgentBench (MAB; n = 3,671), and LongMemEval (n = 500). The reference run records 84.87% on LoCoMo factoid categories and 68.61% on adversarial abstention, 61.49% on MAB across two repetitions,
arXiv:2607.18966v1 Announce Type: cross Abstract: Language models trained with reinforcement learning may learn to optimize the grader's judgment rather than the intended objective. This "reward-seeking" is difficult to measure because a model that pursues the grader's judgment and one that pursues the intended objective behave identically whenever the grader rewards the intended behavior. We measure reward-seeking using Contrastive Synthetic Document Finetuning to change a model's beliefs about what the grader rewards, putting those beliefs in conflict with what users or developers want, and measuring the rate at which the model adopts each party's preferred behavior. Applied to intermediate checkpoints of a capabilities-focused OpenAI o3 RL run, without safety training, we find that these checkpoints often side with grader preferences over those of users or developers on coding and alignment tasks. This tendency to side with the grader trends upward throughout RL training. For exampl
arXiv:2607.18955v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards (RLVR) has substantially improved the reasoning capabilities of large language models on tasks such as mathematical reasoning and code generation. However, most RLVR methods assign a scalar outcome reward to an entire trajectory, resulting in sparse supervision and limited token-level credit assignment. On-policy distillation (OPD) provides denser supervision by distilling token-level distributions from a stronger teacher model, but requires an additional teacher and typically assumes a shared vocabulary. On-policy self-distillation (OPSD) removes this dependency by conditioning the same model on privileged information to construct a teacher policy. However, directly matching the teacher distribution may cause information leakage and unstable optimization. RLSD avoids direct matching by using the teacher signal only to modulate update magnitudes, but it cannot provide an explicit correction
arXiv:2607.18867v1 Announce Type: cross Abstract: Large language models leak parametric knowledge of realized outcomes into historical financial decision tasks. Existence is settled; what users lack is a cheap way to audit a given model for it. We present HindsightBench, a black-box behavioral audit protocol that profiles parametric hindsight in any time-indexed LLM decision task at probe-level cost (no backtests, no logprobs, no corpus access). The protocol chains a four-arm date-manipulation matrix (revealed/date-only/masked/transplanted), dual memory probes (date recovery; outcome recall), and six per-model metrics -- trigger strength, transplant effect, post-cutoff placebo, recoverability, behaviorally effective knowledge cutoff, and a recall-accuracy dissociation coefficient -- with explicit gates where identifiability is data-dependent. Applying it to 15 models from seven vendors on a 258-node vintage-correct macro panel yields three headline patterns: (i) the date-trigger reflex
arXiv:2607.18806v1 Announce Type: cross Abstract: This paper proposes AI Tour Meeting, a group travel planning framework powered by multiple Large Language Model (LLM)-based agents. The agents are instantiated with distinct personas and collaboratively seek an itinerary that satisfies their constraints and preferences through natural language discussion. The framework enables easy and flexible orchestration of such discussions by providing interfaces for configuring agent personas, discussion workflows, monitoring, and LLM deployment. Its primary use case is a simulation tool for analyzing the behavior of multiple LLM agents during tour planning discussions. This paper demonstrates the utility of the framework by presenting system validation and several analytical results obtained by the framework.
arXiv:2607.18795v1 Announce Type: cross Abstract: Researchers, managers and policymakers are exploring different approaches and data sources to map the development and the diffusion of Artificial Intelligence (AI). In this research note, we illustrate the opportunities offered by trademark data. We argue that AI trademarks can complement AI patents to capture different dimensions of AI innovation. AI trademarks can reveal the extent and ways in which companies exploit AI technologies to develop new goods and services. Importantly, trademark data offer a timely and globally available data source that covers all economic sectors. We present insights from using AI trademarks in an empirical exploration of Italian firms. In our discussion, we reflect on how AI trademarks can be used at different levels of analysis to tackle emerging questions about the development and diffusion of AI.
arXiv:2607.18767v1 Announce Type: cross Abstract: The deployment of Small Language Models (SLMs) in educational settings offers significant advantages in terms of privacy, cost, and scalability. However, SLMs often struggle with complex vision-based tasks, such as grading handwritten student exams, due to the high computational cost of processing large images and the visual distractions present on a full page. In this paper, we investigate whether cropping student responses using bounding boxes can improve the accuracy and computational efficiency of SLMs on a short-answer grading task. Using a dataset of scanned handwritten responses from the 2025 Australian Physics Olympiad, we evaluate the performance of several models ranging from 4B to 72B parameters under varying conditions of Chain of Thought (CoT) prompting and image cropping. Our results demonstrate that using bounding boxes significantly improves grading accuracy and reduces computational cost (FLOPs) across models. We conclu
arXiv:2607.18756v1 Announce Type: cross Abstract: Public institutions hold large volumes of sensitive documents and support tickets that cannot leave the premises, ruling out cloud-hosted language models entirely. We report on RAGAL, a retrieval-augmented assistant for the technical-support team of AFIR, the Romanian Agency for Financing Rural Investments, built and operated under three hard constraints: zero data egress (no external API calls, even for synthetic data), a read-only mandate (the assistant drafts, humans execute), and a single 8 GB consumer laptop as the only development and training machine. Over a Romanian-language corpus of ~25,000 chunks -- 15,073 resolved support tickets and internal normative documents -- we show that the highest-leverage investments were retrieval engineering and retriever fine-tuning rather than a larger generator: hybrid dense-sparse retrieval with intent routing raised our internal evaluation from 62% to 81%, and fine-tuning the bge-m3 embedder
arXiv:2607.18754v1 Announce Type: cross Abstract: LLM agent failures are difficult to debug because the step where an error surfaces is often not the one that caused it. Existing observability tools replay execution traces but provide little support for identifying the root cause or translating diagnosis into recovery. We present AgentDebugX, an open-source debugging framework that organizes debugging as a closed loop of Detect, Attribute, Recover, and Rerun. At its core, DeepDebug performs multi-turn root-cause diagnosis through global trajectory understanding, structure-guided investigation, and cross-examination. On the Who and When benchmark, DeepDebug achieves the best strict attribution accuracy among the evaluated methods on both tested open-weight backbones, reaching 28.8 percent exact agent-and-step accuracy on qwen3.5-9b versus 21.7 percent for the strongest single-pass baseline. On GAIA, DeepDebug repairs 13 of 73 failed tasks in a single rerun, compared with 4 to 6 for thre
arXiv:2607.18749v1 Announce Type: cross Abstract: Translating brain signals into text could restore communication for people with severe paralysis, yet practically usable systems to date rely on invasive electrocorticography (ECoG). Electroencephalography (EEG) offers a non-invasive alternative, and EEG-to-text (EEG2Text) has been widely explored. Interestingly, however, EEG2Text models generally rely on teacher-forcing evaluation; without it, they fail to generate meaningful decoding. This reliance prevents EEG2Text from being applied in real-world, non-academic settings. This has fueled numerous debates about whether EEG2Text is a meaningful direction, by extension, and whether EEG truly contains decodable linguistic information. Here, using a neuropsychology-informed paradigm, we find that existing EEG2Text benchmarks have neglected EEG instability, a flaw that has confounded inference and sparked debate. Our experiments furnish key evidence for the feasibility of teacher-forcing-fr
arXiv:2607.18722v1 Announce Type: cross Abstract: Asynchronous reinforcement learning improves throughput by decoupling rollout generation from optimization, but staleness is an inevitable byproduct compounded by policy lag, engine delays, and mixture-of-experts routing. From a trust-region perspective, this mismatch is critical: training-inference divergence governs approximation error in finite-horizon bounds, whereas PPO clipping only gates sampled outward updates, acting as a sampled surrogate rather than a full-policy constraint. As a result, high-staleness updates remain weakly controlled in the asynchronous regime where stale rollouts matter most. We introduce the Staleness-Adaptive Trust Region (SAT), which uses the detached sampled log-ratio as a practical staleness proxy, identifies high-mismatch tails within each batch via staleness-based kernel scaling, and contracts only the sign-selected endpoint of the nominal PPO interval. This preserves baseline behavior on ordinary to
arXiv:2607.18691v1 Announce Type: cross Abstract: Progresses have been made on understanding emotion mechanisms of large language models (LLMs). However, how to explain emotion in LLMs, or even what constitutes good explanations, are less clear. Emotion representations, components, circuits are widely recoverable, but as explanations of a model's own computation they are circular; the emotion space dimensions tend to be arbitrary and non-terminating. A pressing question to ask is whether a more primitive set of internal variables does the work: the semantic primes of the Natural Semantic Metalanguage (NSM). Across four instruction-tuned LLMs (Llama-1B, Gemma-2B, Gemma-9B, OLMo-7B), experiments show that the NSM primes are (1) recoverable internal elements; and (2) on the reference model, intervening with a prime based direction controls emotion about three times as strongly, and twice as selectively, as the best appraisal based direction; and (3) the model treats a prime based explicat
arXiv:2607.18662v1 Announce Type: cross Abstract: We present a practical recipe for building a compact Hindi text-to-speech (TTS) model by distilling a large flow-matching teacher (IndicF5, 337M-parameter DiT) under a severe data budget (~17.6 hours). Training a small model from scratch on this much data fails outright. Instead we warm-start the student from the teacher by pruning depth only: keeping the teacher's width, text dimension, attention heads, and mel/text I/O fixed so all non-block tensors copy one-to-one, and retaining an evenly-spaced subset of transformer blocks. We first measure how much depth the teacher tolerates (it remains near-functional at -27% blocks but collapses past -50%), then descend gradually (22 -> 16 -> 12 -> 8 -> 6 blocks), re-fine-tuning after each prune, with each step gated by an objective ASR word-error-rate (WER) check. The resulting students reach WER 0.00 on unseen sentences at 249M and 190M parameters, and remain robust down to 131M; at 102M we ob
arXiv:2607.18639v1 Announce Type: cross Abstract: Safety interventions on dual-use knowledge typically choose between destroying hazardous content (e.g., unlearning, filtering) and suppressing it at the output layer (e.g., refusal training); both pay a tax in adjacent-domain competence or over-refusal. We argue that the right operation is conditioning, not reduction: we show that hazardous knowledge can be retained in the model and behaviorally gated by a privileged control token. Our method, Token Inoculation, introduces a binding-and-branching approach. First, during continued pre-training, we mark hazardous content by inserting a special token alongside dual-use documents, so the model binds the marker to the underlying semantics of the hazardous domain. Second, during supervised fine-tuning, we teach the model to answer hazardous queries correctly when the special token is present and to refuse them when it is absent, thereby enabling selective refusal without removing dual-use kno
arXiv:2607.18626v1 Announce Type: cross Abstract: Multi-vector dense retrieval models, such as ColBERT, achieve strong retrieval effectiveness by modelling fine-grained token-level interactions between queries and documents. Methods such as PLAID use centroid-based quantisation of each token's vector to reduce the index size and speed up retrieval while maintaining strong effectiveness. In this work, we introduce PLAID-PRF, a method that performs Pseudo-Relevance Feedback (PRF) over PLAID to reformulate ColBERT's query vectors based on the top-retrieved results. In contrast with prior methods that perform PRF on multi-vector retrieval models, PLAID-PRF keeps computational costs low by leveraging the internal PLAID centroid vectors, treating them similarly to tokens in traditional PRF methods. The method selects a small and diverse set of high-utility expansion vectors and appends them to the original query, rerunning PLAID to refine both candidate generation and final scoring. Extensiv
arXiv:2607.18603v1 Announce Type: cross Abstract: We present AutoIndex, a framework for learning representation programs: executable transformations that map raw documents into the representations exposed to a retrieval system. Rather than tuning retrievers, rerankers, or a small set of preprocessing hyperparameters, AutoIndex searches over programs that slice, enrich, normalize, reweight, or reorganize documents before indexing. At each iteration, AutoIndex performs validation-guided program search, in which agents diagnose failures of the current program and synthesize candidate updates, retaining only updates that improve retrieval quality under the resulting index. We evaluate AutoIndex on CRUMB, a benchmark of heterogeneous retrieval tasks, with BM25 held fixed across all experiments. The learned programs improve recall over a static full-document BM25 baseline on all 8 tasks, with average gains of +8.4% in Recall@100 and +8.3% in nDCG@10, and largest gains of +30.5% in Recall@100
arXiv:2607.18451v1 Announce Type: cross Abstract: Frozen encoders are chosen by how well a lightweight head reads a finding from their features, not whether the geometry separates it. Nearest-neighbor discordance does, but with unequal banks the opposite-label neighbor wins on density, not geometry, so prevalence alone makes an uninformed encoder look blind. We introduce CANDOR, a discordance measure whose equal-size banks are symmetric under a label swap, fixing its chance level at exactly one half. Across 22 encoders, 20 datasets from 7 domains, and 605,443 images, this correction reverses the conclusion. Collapse falls below chance almost everywhere, so no encoder is blind, yet all are weak: the best chest model reads pneumothorax at 84.5 AUROC and still places 18.4% of those positives nearer an opposite-label film than its own kind in the same hospital. The same encoder that resolves bird species at 4.5 leaves chest findings at 42.8 and glaucoma at 49.8, at chance and worse than ra