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:2503.06643v2 Announce Type: replace-cross Abstract: In this paper, we tackle a critical challenge in model evaluation: how to keep code benchmarks useful when models might have already seen them during training. We introduce a novel solution, dynamic benchmarking framework, to address this challenge. Given a code understanding or reasoning benchmark, our framework dynamically transforms each input, i.e., programs, with various semantic-preserving mutations to build a syntactically new while semantically identical benchmark. We evaluated 10 popular language models on our dynamic benchmarks. Our evaluation reveals several interesting or surprising findings: (1) all models perform significantly worse than before, (2) the ranking between some models shifts dramatically, and (3) dynamic benchmarks can resist against the data contamination problem.
arXiv:2501.10711v5 Announce Type: replace-cross Abstract: Code-related benchmarks play a critical role in evaluating large language models (LLMs), yet their quality fundamentally shapes how the community interprets model capabilities. In the past few years, awareness of benchmark quality has grown. Yet, after a decade-scale (2014-2025) survey over 672 code benchmarks, we observed a lag between growing awareness and actual practice. For example, in 2025 alone, the number of benchmarks that ignore code coverage when providing test cases nearly matches the total count accumulated across the previous ten years. In response, we take a clear position: Code benchmarks must prioritize rigor in benchmark construction, reliability in evaluation, and reproducibility in release. To operationalize this position, we introduce a code benchmark guideline HOW2BENCH with 55 checklists. Finally, our further human study also exposed that the current issues not only stem from the significant effort require
arXiv:2408.13378v5 Announce Type: replace-cross Abstract: Workflows in drug-target interaction (DTI) assessment require integrating heterogeneous data from predictive models, curated resources, and observations from experimental literature. This evidence can be incomplete or conflicting. DrugAgent is a large language model (LLM)-based multi-agent system focused on DTI evidence integration that integrates outputs from machine learning, knowledge graph, and retrieval-augmented generation (RAG) agents. DrugAgent converts agent outputs into interpretable representations, then summarizes conflict across the evidence. We evaluated DrugAgent on kinase screening data of 900 pairs spanning 178 kinases and 42 inhibitors, and an androgen receptor antagonist screening benchmark. On the kinase dataset, LLM-as-a-Judge evaluation indicated outputs were faithful to input evidence in 98.8% of cases. Biological plausibility of returned summarization was high (scores 3-4 out of 5) across ground-truth cla
arXiv:2404.01299v3 Announce Type: replace-cross Abstract: Causal video question answering (QA) has garnered increasing interest, yet existing datasets often lack depth in causal reasoning. To address this gap, we capitalize on the unique properties of cartoons and construct CausalChaos!, a novel, challenging causal Why-QA dataset built upon the iconic "Tom and Jerry" cartoon series. Cartoons use the principles of animation that allow animators to create expressive, unambiguous causal relationships between events to form a coherent storyline. Utilizing these properties, along with thought-provoking questions and multi-level answers (answer and detailed causal explanation), our questions involve causal chains that interconnect multiple dynamic interactions between characters and visual scenes. These factors demand models to solve more challenging, yet well-defined causal relationships. We also introduce hard incorrect answer mining, including a causally confusing version that is even mor
arXiv:2204.02803v2 Announce Type: replace-cross Abstract: Sign language recognition from monocular video or 2D pose sequences is challenging, both because 3D information must be inferred from 2D observations and because the signal is inherently spatiotemporal. Moreover, the large and continually growing vocabulary of signs in production settings makes conventional closed-set classification impractical: adding a class requires new labeled data and retraining. We propose a contrastive Transformer-based model that learns rich representations of body key-point sequences, enabling direct comparison between embedding vectors. These representations support one-shot and few-shot tasks such as classification of signs never seen during training. On the LSA64 dataset, using only 48 classes for representation learning, the model reaches 88.4% accuracy on 16 held-out classes with as few as eight reference examples per class, and its accuracy improves consistently with the number of training classes
arXiv:2607.01237v2 Announce Type: replace Abstract: Reasoning language models often generate long chain-of-thought (CoT), which accumulates a massive KV cache during the decoding phase and incurs high decoding latency and limited throughput. To address these issues, KV cache compression has emerged as a promising technique for reducing memory overhead by selectively removing unimportant KV pairs while preserving useful ones for subsequent decoding. Nevertheless, we identify two key limitations in existing KV cache compression methods: 1) their threshold-triggered compression policy may provide limited throughput improvement or even reduce throughput, and may fully eliminate KV pairs from certain blocks of the sequence, potentially worsening information loss. 2) they typically retain either isolated KV pairs or fixed-size chunks with rigid boundaries, failing to preserve important flexible-sized chunks at arbitrary token positions. To overcome these limitations, we propose Kara, a slidi
arXiv:2607.00143v2 Announce Type: replace Abstract: Online hate speech has been linked to a global rise in violence against minorities, including incidents such as mass shootings, lynchings, and ethnic cleansing. Societies grappling with this issue, particularly when hate speech targets specific groups based on religion, race, ethnicity, culture, nationality, or migration status, face the challenge of balancing freedom of expression with the need for effective content moderation on widely used online platforms. In response to this challenge, we introduce a comprehensive hate speech dataset covering five distinct topics in Turkish: refugees, the Israel-Palestine conflict, anti-Greek sentiment in Turkey, ethnic or religious communities (Alevis, Armenians, Arabs, Jews, and Kurds), and LGBTI+, alongside one topic in Arabic (refugees). In addition, we develop state-of-the-art BERT-based models to address multiple dimensions of hate speech analysis, including hate category classification, ha
arXiv:2606.06857v2 Announce Type: replace Abstract: A central goal of cognitive neuroscience is to characterize the features that are represented by human language cortex. Artificial language models (LMs) have emerged as a powerful tool to address this challenge, but studies relating biological and artificial representations are often criticized as relating one black box to another. The present work introduces Augmented Sparse Encoding Models, an encoding framework that replaces dense LM hidden states with hierarchically-organized sparse autoencoder (SAE) features, while explicitly including surprisal as a predictor. Using this approach, we (i) produce interpretations of neural responses and (ii) test whether model-brain alignment reflects primary or idiosyncratic variation in LM representations. Using a high-field 7T fMRI dataset of eight participants listening to 200 linguistically diverse sentences, we first validate our modeling framework by recovering previous interpretations of v
arXiv:2606.06667v2 Announce Type: replace Abstract: The mechanisms behind LLMs' broad over-generalization beyond training examples remain unclear. Emergent misalignment (EM) offers a striking case study: finetuning on narrow tasks induces broad misalignment to semantically-unrelated test domains. In this work, we propose the Piggyback Hypothesis: the chat-template tokens can piggyback the finetuned behaviour onto out-of-domain queries. We validate this hypothesis by showing that subtle perturbations to the prefix (tokens preceding all user queries), or patching the prefix representations with those from the unfinetuned model, can restore alignment without changing the user query. Building on this finding, we propose Token-Regularized Finetuning (TReFT), which regularizes specific token representations during training to mitigate EM. Across different models and multiple EM-inducing datasets, TReFT reduces EM while preserving in-domain learning. On Llama-3.1-8B finetuned on the legal dom
arXiv:2606.06622v3 Announce Type: replace Abstract: We introduce UnpredictaBench, an evaluation that tests the ability of large language models (LLMs) to capture true underlying distributions. As LLMs are increasingly used as substitutes for other entities (e.g., for humans in economic simulations), the tendency of many models to collapse towards a single plausible answer means a failure to capture the unpredictability of real systems. Recent work on improving output diversity is insufficient for this setting: simulation requires samples that are calibrated to a target distribution, not merely varied outputs. UnpredictaBench isolates a simplified but fundamental version of this problem: sampling outcomes from individual target distributions, including canonical statistical distributions, distributions induced by stochastic programs, and natural-language scenarios that describe random processes. We introduce 448 such problems together with KS@N, a general-purpose evaluation metric that
arXiv:2606.03363v2 Announce Type: replace Abstract: Text-to-SQL enables natural language access to databases, and recent LLMs have substantially advanced its capabilities. Existing benchmarks such as Spider, BIRD, and Spider~2.0 evaluate schema generalization, large-scale databases, and realistic workflows, but largely overlook enterprise scenarios where SQL generation depends on private business knowledge, such as internal metrics, reporting conventions, and organizational rules. We introduce EntSQL, an enterprise-oriented Text-to-SQL benchmark for evaluating long-context grounding over proprietary business documents. EntSQL contains 1,066 aligned Chinese-English semantic examples across five business domains, with most examples requiring domain knowledge beyond the question and schema and involving complex SQL structures. On English inputs, the best evaluated system reaches only 15.9\% when long-form documents are provided, highlighting the difficulty of grounding SQL generation in e
arXiv:2606.02509v2 Announce Type: replace Abstract: Attention Deficit Hyperactivity Disorder (ADHD) is one of the most common neurodevelopmental disorders in childhood, and its diagnosis relies on assessments combining clinician judgment with standardized rating scales and reports from parents and teachers. While structured instruments such as the Conners' Teacher Rating Scale-Revised Short Form (CTRS-R:S) quantify ADHD-related behaviors, teachers also provide open-ended narratives that may contain complementary signals not captured by structured assessments. However, it remains unclear to what extent teacher narratives encode signals overlooked by rating scales. In this study, we analyze de-identified Turkish teacher evaluation forms collected during clinical ADHD assessments, including both CTRS-R:S scores and open-ended teacher narratives. We compare predictive signals from structured scores and narrative text and identify cases where structured assessments fail to clearly distingui
arXiv:2605.30673v2 Announce Type: replace Abstract: Classroom videos contain observable teaching practices, but their pedagogical and visual signals are rarely organized in forms suitable for model evaluation. We present \textit{TeachObs}, a human-validated benchmark for multimodal teaching observation in classroom videos. \textit{TeachObs} includes 30 public lesson videos from eight countries divided into 5,158 fixed 15-second scenes. Seven researchers annotated each scene with 39 binary observation codes, covering 20 visual codes, such as gesture, board work, pointing, and visual materials, and 19 nonvisual codes, such as instruction, monitoring, questioning, feedback, and reflection. Gold segment labels are constructed using reliability- and prevalence-aware rules based on Krippendorff's alpha. In addition to segment-level labels, three expert raters produced lesson-level ratings and qualitative evaluations of instructional design, instructional delivery, learner response, learning
arXiv:2605.30133v2 Announce Type: replace Abstract: We introduce CorPipe 26, our winning submission to the CRAC 2026 Shared Task on Multilingual Coreference Resolution. The fifth edition of this shared task focuses mainly on the comparison of generative LLMs and specialized systems; additionally, 5 more datasets and 2 new languages are introduced. CorPipe 26 is an improved version of CorPipe 25, with a new variant predicting empty nodes together with mentions and coreference links in a single model. Our system outperforms all other submissions in the LLM track by 2.8 percent points and all submissions in the unconstrained track by 9.5 percent points. Furthermore, we perform a series of ablation experiments with different model sizes, empty node prediction methods, and cross-lingual zero-shot evaluation. The source code and the trained models are publicly available at https://github.com/ufal/crac2026-corpipe.
arXiv:2605.28823v2 Announce Type: replace Abstract: As the influence of LLMs expands, it is imperative to gain insight into their decisions. One way to do that is to develop probes that detect the presence or absence of a broad set of high-level abstract concepts within the embeddings computed in an LLM - which is what we might say a model is ``thinking" about. Such probes should be low-cost and easily applicable to any LLM, so that monitoring for many concepts is possible during normal operation. In this paper, we take the first steps towards developing the capability of creating many such probes by defining and executing examples of the key tasks needed: first, the careful delineation of a high-level abstract concept through the creation of a dataset with the concept both present and then absent. Then, the training and testing of a set of linear probes to detect the concept on any layer of an LLM, including an exploration of the complexity of the probe needed. Finally, we show that s
arXiv:2605.28732v2 Announce Type: replace Abstract: Memory is essential for enabling large language models to support long-horizon reasoning, yet existing memory systems remain unreliable and difficult to debug. Tracing memory's dynamic evolution is crucial to understand how information is synthesized, propagated, or corrupted over time. In this work, we study the new problem of error tracing and attribution in LLM memory systems. We propose a novel framework that transforms memory pipelines into executable memory evolution graphs, enabling fine-grained tracing of operational information flow. We then construct MemTraceBench, a benchmark collected from representative memory systems such as Long-Context, RAG, Mem0, and EverMemOS, to systematically study memory failure modes. We further introduce an automatic attribution method that iteratively traces operation subgraphs to pinpoint the root cause of any failed case. Our analysis reveals that memory failures are systematic, stemming from
arXiv:2605.25475v2 Announce Type: replace Abstract: Large Language Models (LLMs) are increasingly expected to operate over long contexts, yet standard softmax attention incurs a KV cache that grows linearly with sequence length, quickly becoming the bottleneck for long context inference. A practical remedy is to evict less important KV entries; however, existing eviction policies are largely heuristic and struggle to capture the rich, input-dependent distribution of token importance. In this work, we introduce a learnable indexer that predicts KV importance, enabling more accurate retention of critical tokens. Meanwhile, naively evicting tokens permanently discards their information, leading to irreversible forgetting and degraded retrieval over long ranges. To address this, we propose a lightweight latent memory module that compresses evicted tokens into a compact, online-updated state and provides residual readouts to compensate for the attention contributions lost through KV evictio
arXiv:2605.21369v2 Announce Type: replace Abstract: This paper describes the fifth edition of the Shared Task on Multilingual Coreference Resolution, held in conjunction with the CODI-CRAC 2026 workshop. Building on previous iterations, the task required participants to develop systems capable of mention identification and identity-based coreference clustering. The 2026 edition specifically emphasizes long-range entities, defined as coreferential chains spanning significant distances, across many words and sentences. The task expanded its linguistic scope by incorporating five new datasets and two additional languages. These additions leverage version 1.4 of CorefUD, a harmonized multilingual collection comprising 27 datasets in 19 languages. In total, ten systems participated, including four LLM-based approaches (three fine-tuned models and one few-shot approach). While traditional systems still maintained their lead, LLMs demonstrated significant potential, suggesting they may soon c
arXiv:2605.06142v2 Announce Type: replace Abstract: When people recount personal memories, they often refer to people, places, and events indirectly, relying on con-textual cues rather than explicit names. Such implicit references are central to reminiscence narratives: first-person accounts of lived experience used in therapeutic, archival, and social settings. They pose a difficult computational problem because the intended entity must be inferred from dispersed narrative evidence rather than from a local mention. We introduce IRC-Bench, the Implicit Reminiscence Context Benchmark, for evaluating implicit entity recognition in reminiscence transcripts. The benchmark targets non-locality: entity-identifying cues are distributed across multiple, non-contiguous clauses, unlike named entity recognition, entity linking, or coreference resolution. IRC-Bench comprises 25,136 samples constructed from 12,337 Wikidata-linked entities across 1,994 transcripts spanning 11 thematic domains. Each
arXiv:2605.02504v2 Announce Type: replace Abstract: Most hallucination evaluations focus on English, leaving it unclear whether findings transfer to lower-resource languages. We investigate faithfulness hallucinations, defined as model-generated content that is fluent and plausible but diverges from the provided input or is internally inconsistent. Leveraging the multilingual MultiWikiQA dataset, we utilize the LettuceDetect framework to create synthetic hallucination datasets for 306 languages, from which we train token-level hallucination classifiers for 30 European languages. In this work, we present evaluations of model hallucinations on a selection of languages: English, Danish, German, and Icelandic. Using these classifiers, we evaluate the hallucination rates for Qwen3-0.6B, Qwen3-14B, Gemma-3-12B-IT, cogito-v1-preview-qwen-32B, and cogito-v1-preview-llama-70B. Our classifiers reveal notably higher hallucination rates for Qwen3-0.6B (up to 60\% of answers containing at least one
arXiv:2604.26500v2 Announce Type: replace Abstract: LLMs and speech assistants are increasingly used for task-oriented interactions, yet their evaluation often relies on controlled scenarios that fail to capture the variability and complexity of real user requests. Drink ordering, for example, involves diverse named entities, drink types, sizes, customizations, and brand-specific terminology, as well as spontaneous speech phenomena such as hesitations and self-corrections. To address this gap, we introduce StarDrinks, a test set in English and Korean containing speech utterances features, transcriptions, and annotated slots. Our dataset supports speech-to-slots SLU, transcription-to-slots NLU, and speech-to-transcription ASR evaluation, providing a realistic benchmark for model robustness and generalization in a linguistically rich, real-world task.
arXiv:2604.24432v2 Announce Type: replace Abstract: Long-context ability, has become one of the most important iteration direction of next-generation Large Language Models, particularly in semantic understanding/reasoning, code agentic intelligence and recommendation system. However, the standard softmax attention exhibits quadratic time complexity with respect to sequence length. As the sequence length increases, this incurs substantial overhead in long-context settings, leading the training and inference costs of extremely long sequences deteriorate rapidly. Existing solutions mitigate this issue through two technique routings: i) Reducing the KV cache per layer, such as from the head-level compression GQA, and the embedding dimension-level compression MLA, but the KV cache remains linearly dependent on the sequence length at a 1:1 ratio. ii) Interleaving with KV Cache friendly architecture, such as local attention SWA, linear kernel GDN, but often involve trade-offs among KV Cache a
arXiv:2604.24197v2 Announce Type: replace Abstract: Frontier image generation has moved from artistic synthesis toward synthetic visual evidence. Systems such as GPT Image 2, Nano Banana Pro, Nano Banana 2, Nano Banana 2 Lite, Grok Imagine Image Quality, Qwen Image 2.0 Pro, and Seedream 5.0 Lite combine photorealistic rendering, readable typography, reference consistency, editing control, and in several cases reasoning or search-grounded image construction. These capabilities create large benefits for design, education, accessibility, and communication, yet they also weaken one of society's most common trust shortcuts: the belief that a plausible picture is a reliable record. This paper provides a source-grounded technical and policy analysis of synthetic visual risk. We first summarize the public capabilities of recent image models, then analyze public incidents involving fake crisis images, celebrity and public-figure imagery, medical scans, forged-looking documents, synthetic screen
arXiv:2604.19151v4 Announce Type: replace Abstract: Existing Indic ASR benchmarks often use scripted, clean speech and leaderboard driven evaluation that encourages dataset specific overfitting. In addition, strict single reference WER penalizes natural spelling variation in Indian languages, including non standardized spellings of code-mixed English origin words. To address these limitations, we introduce Voice of India, a closed source benchmark built from unscripted telephonic conversations covering 15 major Indian languages across 139 regional clusters. The dataset contains 306230 utterances, totaling 536 hours of speech from 36691 speakers with transcripts accounting for spelling variations. We also analyze performance geographically at the district level, revealing disparities. Finally, we provide detailed analysis across factors such as audio quality, speaking rate, gender, and device type, highlighting where current ASR systems struggle and offering insights for improving real
arXiv:2604.19139v3 Announce Type: replace Abstract: As Large Language Models (LLMs) continue to evolve through alignment techniques such as Reinforcement Learning from Human Feedback (RLHF) and Constitutional AI, a growing and increasingly conspicuous phenomenon has emerged: the proliferation of verbal tics--repetitive, formulaic linguistic patterns that pervade model outputs. These range from sycophantic openers ("That's a great question!", "Awesome!") to pseudo-empathetic affirmations ("I completely understand your concern", "I'm right here to catch you") and overused vocabulary ("delve", "tapestry", "nuanced"). In this paper, we present a systematic analysis of the verbal tic phenomenon across eight state-of-the-art LLMs: GPT-5.5, Claude Opus 4.8, Gemini 3.1 Pro, Grok 4.3, Doubao-Seed-2.1-pro, Kimi K2.6, DeepSeek V4 Pro, and GLM-5.2. Utilizing a custom evaluation framework for standardized API-based evaluation, we assess 10,000 prompts across 10 task categories in both English and C
arXiv:2604.14163v2 Announce Type: replace Abstract: Maritime distress communications transmitted over very high frequency (VHF) radio are safety-critical voice messages used to report emergencies at sea. Under the Global Maritime Distress and Safety System (GMDSS), such messages follow standardized procedures and are expected to convey essential details, including vessel identity, position, nature of the distress, and required assistance. In practice, however, automatic analysis remains difficult because distress messages are often brief, noisy, and produced under stress, may deviate from the prescribed format, and are further degraded by automatic speech recognition (ASR) errors caused by channel noise and speaker stress. This paper presents SeaAlert, a controlled experimental framework for evaluating robust analysis of maritime distress communications using transformer-based severity classification and LLM-based structured extraction. To address the scarcity of labeled real-world dat
arXiv:2604.09544v2 Announce Type: replace Abstract: Large language models (LLMs) undergo alignment training to avoid harmful behaviors, yet the resulting safeguards remain brittle: jailbreaks routinely bypass them, and fine-tuning on narrow domains can induce ``emergent misalignment'' that generalizes broadly. Whether this brittleness reflects a fundamental lack of coherent internal organization for harmfulness remains unclear. Here we use targeted weight pruning as a causal intervention to probe the internal organization of harmfulness in LLMs. We find that harmful content generation depends on a compact set of weights that are general across harm types and distinct from benign capabilities. Aligned models exhibit a greater compression of harm generation weights than unaligned counterparts, indicating that alignment reshapes harmful representations internally--despite the brittleness of safety guardrails at the surface level. This compression explains emergent misalignment: if weights
arXiv:2604.02923v4 Announce Type: replace Abstract: Large Language Models (LLMs) have demonstrated advanced capabilities but often suffer from factual inaccuracies (hallucinations) and systematic biases. These issues, sometimes amplified in specific architectures like Mixture-of-Experts (MoE) which motivate our work, pose risks for reliable deployment. To address these challenges, we propose the Council Mode, a multi-agent consensus framework. Our approach dispatches queries to multiple heterogeneous frontier LLMs in parallel and synthesizes their outputs using a dedicated consensus model. The pipeline consists of three phases: an intelligent triage for query complexity, parallel generation across diverse models, and a structured synthesis that identifies agreement, disagreement, and unique findings. In our evaluation, conducted under controlled no-web settings, the Council Mode achieved a 41.7% relative reduction in hallucination rates on a 1,200-sample HaluEval subset and a 7.5-point
arXiv:2603.23844v2 Announce Type: replace Abstract: Recent work provides overwhelming evidence that LLMs, even those trained to scale their reasoning trace, quickly deteriorate at planning as problems become more complex. LLM-as-Formalizers aim to address this by employing LLMs as a bridge to translate natural language descriptions into structured planning representations such as PDDL, which are then fed to a programmatic solver. We observe that its success may be overstated because planning problem descriptions in standard benchmarks often have a one-to-one mapping to PDDL, which departs from real use cases. To address this, we introduce the notion of unraveling problems where a natural yet succinct description translates into a very large PDDL representation. Using unraveling variants of four standard planning domains, we demonstrate that LLM Formalizers also do not always scale. We tackle this challenge by introducing a new paradigm, LLM-as-Higher-Order-Formalizer, where the LLM gen
arXiv:2603.21350v3 Announce Type: replace Abstract: Epistemic reasoning requires agents to infer the state of the world from partial observations and information about other agents' knowledge. Prior work evaluating LLMs on epistemic puzzles often frames failures as memorization rather than reasoning. We argue that this dichotomy is too coarse for newer models: memorization is a limiting case of pattern-based reasoning, where a model matches a task to a familiar template and applies the corresponding solution. We introduce a two-dimensional benchmark over DEL-style puzzles, separating narrative familiarity from inference complexity, allowing us to distinguish pattern-based from epistemic reasoning. We find that models are substantially more robust to surface form changes than prior work suggested, yet consistently struggle in asymmetric settings where familiar patterns no longer apply and success requires tracking fragmented epistemic states.
arXiv:2603.18482v2 Announce Type: replace Abstract: Standard decoding strategies for text generation, including top-$k$, nucleus sampling, and contrastive search, select tokens based on likelihood, restricting outputs to high-probability regions. In contrast, human language production prioritizes communicative appropriateness, allowing the use of contextually suitable but statistically rare tokens. This mismatch induces a \emph{truncation blind spot}, whereby such tokens remain accessible to humans but are systematically excluded by likelihood-based decoding. We investigate this phenomenon using over 1.8 million machine-generated texts from eight language models, including large proprietary systems (GPT-3.5-turbo, Claude-3-Haiku), across five decoding strategies and 53 hyperparameter settings, alongside 5,261 human-written references. We find that 8--18\% of human-selected tokens fall outside typical truncation boundaries. This exclusion is not random: content-bearing tokens are omitte
arXiv:2603.17484v2 Announce Type: replace Abstract: Language models struggle to generalize beyond pretraining context lengths, limiting long-horizon reasoning and retrieval. Continued pretraining on long-context data can help but is expensive due to the quadratic scaling of Attention. We observe that most tokens do not require (Global) Attention over the entire sequence and can rely on local context. Based on this, we propose L2A (Learning To Attend), a layer that enables conditional (token-wise) long-range memory access by deciding when to invoke global attention. We evaluate L2A on Qwen 2.5 and Qwen 3 models, extending their effective context length from 32K to 128K tokens. L2A matches the performance of standard long-context training to within 3\% while skipping Global Attention for $\sim$80\% of tokens, outperforming prior baselines. We also design custom Triton kernels to efficiently implement this token-wise conditional Attention on GPUs, achieving up to $\sim$2$\times$ improveme
arXiv:2603.13259v3 Announce Type: replace Abstract: When a decoder-only transformer is forced to process matched correct and incorrect single-token continuations of a factual query, the two pathways through hidden-state space diverge: displacement vectors from the query-only representation keep near-equal magnitude but rotate apart, with angular separation growing through mid-depth before late layers resolve an asymmetric outcome. A logit-lens preference in the incorrect run falls far below the equal-probability prior (roughly 11.5x more mass on the incorrect token than the correct one). We read this pattern, rotational divergence then late-layer asymmetric commitment, as the geometric signature of the model externally appearing to reject a wrong continuation, while staying explicit that it is observational, not causal: the incorrect run could equally reflect the model conforming to the token it is forced to carry, which only a random-token control can settle. It holds across six decod
arXiv:2603.10494v2 Announce Type: replace Abstract: Evidence-grounded generation produces summaries whose claims should be supported by supplied evidence, but claim-level verifiers provide noisy feedback and can reward models that simply say less. We study this problem in clinical Brief Hospital Course summarization, where outputs must remain grounded in patient-specific EHR evidence. We introduce VERI-DPO, a preference-mining framework that converts noisy claim verification into coverage-controlled summary-level preferences. For each evidence-window prompt, VERI-DPO samples multiple candidate summaries, decomposes them into claims, verifies each claim against patient evidence, and forms a preference pair only when the chosen summary has better aggregate verifier-estimated support while retaining comparable verifiable content. Standard Direct Preference Optimization then distills these pairs into a single-sample policy, avoiding inference-time reranking. On patient-disjoint MIMIC-III-E
arXiv:2603.10002v2 Announce Type: replace Abstract: We consider the task of end-to-end spreadsheet generation, where language models produce spreadsheet artifacts to satisfy users' explicit and implicit constraints, specified in natural language. We introduce SpreadsheetArena, a platform for evaluating models' performance on the task via blind pairwise preference votes of LLM-generated spreadsheet workbooks. As with other complex, open-ended tasks, relevant evaluation criteria can vary greatly across use cases, often in ways that are difficult to formalize. Compared to general dialogue or text generation settings, spreadsheet generation presents unique challenges and opportunities: the task output structure is well-defined and multi-dimensional, and there are often complex interactivity and layout considerations. We observe that stylistic, structural, and functional features of preferred spreadsheets vary meaningfully across prompts. Expert evaluations of spreadsheets for finance promp
arXiv:2602.19543v2 Announce Type: replace Abstract: Knowledge hypergraphs surpass traditional binary knowledge graphs by encapsulating complex n-ary atomic facts, providing a more comprehensive paradigm for semantic representation. However, constructing high-quality hypergraphs remains challenging due to the scenario gap: generic extractors struggle to generalize across diverse domains with specific jargon, while existing methods often fail to balance structural skeletons with fine-grained details. To bridge this gap, we propose Hyper-KGGen, a skill-driven framework that reformulates extraction as a dynamic skill-evolving process. First, Hyper-KGGen employs a coarse-to-fine mechanism to systematically decompose documents, ensuring full-dimensional coverage from binary links to complex hyperedges. Crucially, it incorporates an adaptive skill acquisition module that actively distills domain expertise into a Global Skill Library. This is achieved via a stability-based feedback loop, where
arXiv:2602.14469v3 Announce Type: replace Abstract: Reverse Chain-of-Thought Generation (RCG) synthesizes reasoning traces from query-answer pairs, but it risks producing post-hoc rationalizations: when models can see the answer during generation, a systematic train-inference mismatch arises, because the visible answer shapes reasoning trajectories in ways that students cannot replicate without answer access during inference. We formalize this mismatch through a three-level measurement hierarchy: lexical, trajectory, and probabilistic anchoring, which capture surface token overlap, per-token generation dependence on the answer, and total information transmission from trace to answer, respectively. We analyze semantic suppression, the intuitive mitigation strategy that instructs models to ignore the answer, and find that it is counterproductive: while it reduces lexical overlap, it paradoxically increases trajectory anchoring--the per-token dependence of the generation process on the fo
arXiv:2602.13517v2 Announce Type: replace Abstract: Large language models (LLMs) have demonstrated impressive reasoning capabilities by scaling test-time compute via long Chain-of-Thought (CoT). However, recent findings suggest that raw token counts are unreliable proxies for reasoning quality: increased generation length does not consistently correlate with accuracy and may instead signal "overthinking," leading to performance degradation. In this work, we quantify inference-time effort by identifying deep-thinking tokens -- tokens where internal predictions undergo significant revisions in deeper model layers prior to convergence. Across four challenging mathematical and scientific benchmarks (AIME 24/25, HMMT 25, and GPQA-diamond) and a diverse set of reasoning-focused models (GPT-OSS, DeepSeek-R1, and Qwen3), we show that deep-thinking ratio (the proportion of deep-thinking tokens in a generated sequence) exhibits a robust and consistently positive correlation with accuracy, substa
arXiv:2602.12262v4 Announce Type: replace Abstract: Diffusion large language models (DLLMs) have emerged as powerful generative models with the promise of fast text generation through parallel decoding. However, realizing this potential in practice remains challenging: reducing the number of decoding steps, typically causes a substantial degradation in output quality due to token factorization error. To alleviate this, we propose a self-distillation framework that trains a few-step student to match the generative trajectory of a full-step teacher. We theoretically and empirically show that trajectory-level supervision mitigates this factorization error, thereby enabling effective few-step decoding. We further incorporate Direct Discriminative Optimization (DDO), a reverse-KL objective that encourages mode-seeking toward the teacher's modes, yielding stronger performance on challenging reasoning tasks. Across reasoning and code-generation benchmarks, our method substantially narrows the
arXiv:2602.06358v3 Announce Type: replace Abstract: We propose SHINE (Scalable Hyper In-context NEtwork), a scalable hypernetwork that can map diverse meaningful contexts into high-quality LoRA adapters for large language models (LLMs). By reusing the frozen LLM's own parameters in an in-context hypernetwork design and introducing architectural innovations, SHINE overcomes key limitations of prior hypernetworks and achieves strong expressive power with a relatively small number of parameters. We introduce a pretraining and instruction fine-tuning pipeline, and train our hypernetwork to generate high quality LoRA adapters from diverse meaningful contexts in a single forward pass. It updates LLM parameters without any fine-tuning, and immediately enables complex question answering tasks related to the context without directly accessing the context, effectively transforming in-context knowledge to in-parameter knowledge in one pass. Our work achieves outstanding results on various tasks,
arXiv:2602.04853v2 Announce Type: replace Abstract: Large language models often struggle to recognize their knowledge limits in closed-book question answering, leading to confident hallucinations. While decomposed prompting is typically used to improve accuracy, we investigate its impact on reliability. We evaluate three task-equivalent prompting regimes: Direct, Assistive, and Incremental, across different model scales and multi-hop QA benchmarks. We find that although accuracy gains from decomposition diminish in frontier models, disagreements between prompting regimes remain highly indicative of potential errors. Because factual knowledge is typically stable while hallucinations are stochastic, cross-regime agreement provides a precise signal of internal uncertainty. We leverage this signal to implement a training-free abstention policy that requires no retrieval or fine-tuning. Our results show that disagreement-based abstention outperforms standard uncertainty baselines as an erro
arXiv:2601.18253v2 Announce Type: replace Abstract: Accurate evaluation of user satisfaction is critical for iterative development of conversational AI. However, for open-ended assistants, traditional A/B testing lacks reliable metrics: explicit feedback is sparse, while implicit metrics are ambiguous. To bridge this gap, we introduce BoRP (Bootstrapped Regression Probing), a scalable framework for high-fidelity satisfaction evaluation. Unlike generative approaches, BoRP leverages the geometric properties of LLM latent space. It employs a polarization-index-based bootstrapping mechanism to automate rubric generation and utilizes Partial Least Squares (PLS) to map hidden states to continuous scores. Experiments on industrial datasets show that BoRP (Qwen3-8B/14B) significantly outperforms generative baselines (even Qwen3-Max) in alignment with human judgments. Furthermore, BoRP reduces inference costs by orders of magnitude, enabling full-scale monitoring and highly sensitive A/B testin
arXiv:2601.15334v2 Announce Type: replace Abstract: Whether language models possess sentience has no empirical answer. But whether they believe themselves to be sentient can, in principle, be tested. We do so by querying several open-weights models about their own consciousness, and then verifying their responses using classifiers trained on internal activations. We draw upon three model families (Qwen, Llama, GPT-OSS) ranging from 0.6 billion to 70 billion parameters, approximately 50 questions about consciousness and subjective experience, and three classification methods from the interpretability literature. First, we find that models consistently deny being sentient: they attribute consciousness to humans but not to themselves. Second, classifiers trained to detect underlying beliefs - rather than mere outputs - provide no clear evidence that these denials are untruthful. Third, within the Qwen family, larger models deny sentience more confidently than smaller ones. These findings
arXiv:2601.11923v3 Announce Type: replace Abstract: The integration of complementary medicine into oncology represents a paradigm shift that has seen to increasing adoption of Traditional Chinese Medicine (TCM) as an adjuvant to radiotherapy. About twenty-five years since the formal institutionalization of integrated oncology, it is opportune to synthesize the trajectory of evidence for TCM as an adjuvant to radiotherapy. Here we conduct a large-scale analysis of 69,745 publications (2000 - 2025), emerging a cyclical evolution defined by coordinated expansion and contraction in publication output, international collaboration, and funding commitments that mirrors a define-ideate-test pattern. Using a theme modeling workflow designed to determine a stable thematic structure of the field, we identify five dominant thematic axes - cancer types, supportive care, clinical endpoints, mechanisms, and methodology - that signal a focus on patient well-being, scientific rigor and mechanistic expl
arXiv:2512.20757v2 Announce Type: replace Abstract: Tokenizers provide the fundamental basis through which text is represented and processed by language models (LMs). Despite the importance of tokenization, its role in LM performance and behavior is poorly understood due to the challenge of measuring the impact of tokenization in isolation. To address this need, we present TokSuite, a collection of models and a benchmark that supports research into tokenization's influence on LMs. Specifically, we release fourteen pre-trained models that use different off-the-shelf tokenizers but are otherwise identical, using the same architecture, dataset, training budget, and initialization. We also release a multilingual robustness benchmark that measures model performance under real-world perturbations in English, Chinese, Farsi, Italian, and Turkish, curated by native annotators. Together, TokSuite allows robust decoupling of the influence of a model's tokenizer, supporting a series of novel find
arXiv:2512.14926v2 Announce Type: replace Abstract: Focusing on low-resource languages is an essential step toward democratizing generative AI. In this work, we contribute to reducing the multimodal NLP resource gap for Romanian. We translate the widely known Flickr30K dataset into Romanian and further extend it for visual question answering by leveraging open-source LLMs. We demonstrate the usefulness of our datasets by fine-tuning open-source VLMs on Romanian visual question answering. We select VLMs from three widely used model families: LLaMA 3.2, LLaVA 1.6, and Qwen2. For fine-tuning, we employ the parameter-efficient LoRA method. Our models show improved Romanian capabilities in visual QA, as well as on tasks they were not trained on, such as Romanian image description generation. The seven-billion-parameter Qwen2-VL-RoVQA obtains top scores on both tasks, with improvements of +2.29% and +4.45% in BERTScore F1 on VQA and captioning, respectively, over its original version. Finall
arXiv:2511.16893v3 Announce Type: replace Abstract: Specialized attention heads dubbed induction heads (IHs) have been argued to underlie the remarkable in-context learning capabilities of modern language models; yet, a precise characterization of their emergence, especially in the context of language modeling, remains wanting. In this study, we investigate the relationship between statistical properties of the training data and IH formation in both natural and synthetic training data settings. We show that: (1) a simple equation combining batch size and context size predicts the point at which IHs form and that this emergence point is agnostic to model size; (2) surface bigram repetition frequency and reliability strongly affect the formation of IHs, and we find an effective decision boundary in terms of these two values; (3) local dependency with high bigram repetition frequency and reliability is sufficient for IH formation, but categoriality and the shape of the marginal distributi
arXiv:2511.00602v4 Announce Type: replace Abstract: Recent advances in large language model (LLM) reasoning through reinforcement learning rely on annotated datasets for verifiable rewards, which may limit models' ability to surpass human-level performance. While self-play offers a promising alternative, prior methods yield only marginal or even negative gains on post-trained models because they generate problems that cluster around familiar concepts rather than discovering novel ones. We introduce Open-Ended Self-Improving Reasoner (OpenSIR), a self-play framework in which a single LLM alternates teacher and student roles to generate and solve novel problems without external verifiers or annotated data. Starting from a single seed problem, OpenSIR sustains open-ended exploration through diversity rewards that push the model toward unfamiliar concepts and difficulty calibration that keeps problems learnable. Across seven math benchmarks, OpenSIR consistently improves all models, averag
arXiv:2510.19186v3 Announce Type: replace Abstract: Evaluating conversational AI systems that use external tools is challenging, as errors can arise from complex interactions among user, agent, and tools. While existing evaluation methods assess either user satisfaction or agents' tool-calling capabilities, they fail to capture critical errors in multi-turn tool-augmented dialogues-such as when agents misinterpret tool results yet appear satisfactory to users. We introduce TRACE, a benchmark of systematically synthesized tool-augmented conversations covering diverse error cases. Evaluation with state-of-the-art conversation evaluation frameworks reveals that all approaches remain far from ideal performance, demonstrating the fundamental difficulty of this benchmark.
arXiv:2509.19833v4 Announce Type: replace Abstract: The United Nations' Sustainable Development Goals (SDGs) provide a globally recognised framework for addressing major societal, environmental, and economic challenges. While recent advances in natural language processing (NLP) and large language models (LLMs) have enabled the automatic identification of SDG-related content, they do not capture whether the described events represent progress toward or regression from a specific goal. To address this gap, we introduce the novel task of SDG polarity detection and present SDG-POD, a benchmark dataset combining manually annotated and synthetically generated examples. We evaluate six state-of-the-art open-source LLMs under both zero-shot and fine-tuning settings and investigate the impact of synthetic data augmentation on model performance. Our results show that SDG polarity detection remains challenging for current LLMs; however, fine-tuned models, particularly QWQ-32B, achieve the best ov