Papers

I developed a new version of Latent Semantic Scaling (LSS) employing word2vec as a masked language model. Unlike original spatial models, it assigns polarity scores to words and documents as predicted probabilities of seed words to occur in given contexts. These probabilistic polarity scores are more accurate, interpretable and consistent than those spatial polarity models can produce in text analysis. I demonstrate these advantages by applying both probabilistic and spatial models to China Daily's coverage of China and other countries during the coronavirus disease (COVID) pandemic in terms of achievement in health issues. The result suggests that more advanced masked language models would further improve the semisupervised machine learning technique.

This study compares machine learning and deep learning approaches for cyberbullying detection in Indonesian-language Instagram comments. Using a balanced dataset of 650 comments labeled as Bullying and Non-Bullying, the study evaluates Naive Bayes, Logistic Regression, and Support Vector Machine with TF-IDF features, as well as BiLSTM and BiLSTM with Bahdanau Attention. A preprocessing pipeline tailored to informal Indonesian text is applied, including slang normalization, stopword removal, and stemming. The results show that Logistic Regression performs best among the machine learning models, while BiLSTM with Attention achieves the strongest overall deep learning performance. The findings highlight the value of domain-specific preprocessing and show that although deep learning captures contextual patterns more effectively, machine learning remains a competitive option for resource-constrained deployments.

Some text generation tasks, such as Attribute Value Extraction (AVE), require decoding multiple independent sequences from the same document context. While standard autoregressive decoding is slow due to its sequential nature, the independence between output sequences offers an opportunity for parallelism. We present Hyper-Parallel Decoding, a novel decoding algorithm that accelerates offline decoding by leveraging both shared memory and computation across batches. HPD enables out-of-order token generation through position ID manipulation, significantly improving efficiency. Experiments on AVE show that attribute-value pairs are conditionally independent, enabling us to parallelize value generation within each prompt. By further stacking multiple documents within a single prompt, we can decode in parallel up to 96 tokens per prompt. HPD works with all LLMs, and reduces both inference costs and total inference time by up to 13.8X without compromising output quality, potentially saving hundreds of thousands of dollars on industry AVE tasks. Although designed for attribute extraction, HPD makes no assumptions unique to the AVE domain and can in theory be applied to other scenarios with independent output structures.

A predecessor pilot (Cacioli, 2026) found that Llama-3-8B implements prompted sandbagging as positional collapse rather than answer avoidance. However, fixed option ordering in MMLU-Pro left open whether this reflected a model-level position-dominant policy or dataset-level distractor structure. This pre-registered follow-up (3 models, 2,000 MMLU-Pro items, 4 conditions, 24,000 primary trials) added cyclic option-order randomisation as the critical control. The pre-registered item-level same-letter diagnostic did not confirm deterministic position-tracking (same-letter rate 37.3%, below the 50% threshold). However, pre-specified supporting analyses revealed that the response-position distribution under sandbagging was highly stable under complete content rotation (Pearson r = 0.9994; Jensen-Shannon divergence = 0.027, compared to 0.386 between honest and sandbagging conditions). Accuracy spiked to 72.1% when the correct answer coincidentally occupied the preferred position E, and fell to 4.3% at position A. The data provide strong evidence for a soft distributional attractor: under sandbagging instruction, the model enters a low-entropy response-position basin centred on E/F/G that is highly stable and largely content-invariant at the aggregate level. Qwen-2.5-7B served as a negative control (non-compliant, no distributional shift). These results provide evidence, at the 7-9 billion parameter scale, that response-position entropy is a promising black-box behavioural signature of this sandbagging mode.

We introduce Flashback, a reversible string decomposition that repeatedly peels the maximal leading and trailing character runs from a sentinel-wrapped input, recording each pair as one bilateral token. Decomposition and reconstruction both run in O(n) time and space. Our central result is a run-pairing theorem: Flashback is equivalent to pairing the first run of the string with the last, the second with the second-to-last, and so on. This gives an exact token count of 1+[r/2] for a string with r maximal runs, and matches a lower bound that holds for any admissible bilateral run-peeling scheme. From the run-pairing theorem the main structural properties follow as corollaries: the irreducible peeling kernel uses at most two symbols; palindromes are precisely the strings whose run-length encoding is symmetric with an odd number of runs; the image of the decomposition admits an explicit finite-state characterisation; and changing one run length rewrites exactly one content token.

Achieving reliable control of Large Language Models (LLMs) requires a precise, scalable understanding of how they interpret linguistic cues. We introduce a rigorous framework using Shapley values to quantify the steering effect of individual adjectives on model performance, moving beyond anecdotal heuristics to principled attribution. Applying this method to 100 adjectives across a diverse suite of models (including o3, gpt-4o-mini, phi-3, llama-3-70b, and deepseek-r1) on the MMLU benchmark, we uncover several critical findings for AI alignment. First, we find that a small subset of adjectives act as disproportionately powerful "levers," yet their effects are not universal. Cross-model analysis reveals a "family effect": models of a shared lineage exhibit correlated sensitivity profiles, while architecturally distinct models react in a largely uncorrelated manner, challenging the notion of a one-size-fits-all prompting strategy. Second, focused follow-up studies demonstrate that the steering direction of these powerful adjectives is not intrinsic but is highly contingent on their syntactic role and position within the prompt. For larger models like gpt-4o-mini, we provide the first quantitative evidence of strong, non-additive interaction effects where adjectives can synergistically amplify, antagonistically dampen, or even reverse each other's impact. In contrast, smaller models like phi-3 exhibit a more literal and less compositional response. These results suggest that as models scale, their interpretation of prompts becomes more sophisticated but also less predictable, posing a significant challenge for robustly steering model behavior and highlighting the need for compositional and model-specific alignment techniques.

Test-time scaling (TTS) has emerged as a powerful paradigm for improving the reasoning performance of large language models (LLMs) by investing additional compute at inference time. A central component of TTS is the verifier, which selects or scores candidate solutions to guide the search process. While prior work has explored the benefit of verification, a fundamental question remains underexplored: what is the optimal granularity of verification under a given compute budget? Coarse-grained outcome reward models (ORMs) and fine-grained process reward models (PRMs) represent two extremes, yet neither alone achieves compute-optimality across all regimes. In this paper, we establish a unified theoretical framework, called GRACE (\underline{G}ranularity-\underline{R}egulated \underline{A}daptive \underline{C}omputational \underline{E}fficiency), that characterizes the optimal verification granularity as an explicit function of problem difficulty, verifier accuracy, and compute budget. We prove that there exists a phase transition: fine-grained verification dominates when either the compute budget is large or the problem is hard, whereas coarse-grained verification is preferred in the low-budget, easy-problem regime. Our theory unifies Best-of-$N$, beam search, and step-level MCTS within a single Pareto-optimality framework, and motivates an adaptive granularity strategy that provably achieves the compute-performance Pareto frontier. Empirical results on MATH-500, GSM8K, and AIME benchmarks corroborate all four theoretical claims, with our adaptive strategy outperforming fixed-granularity baselines by up to 3.1% accuracy at matched compute.

In-Context Learning (ICL) allows LLMs to adapt to new tasks from a few demonstrations, but its reliability remains a concern: predictions are highly sensitive to both prompt design and the model's ability to understand the context, obscuring whether failures arise from data properties or model limitations. Uncertainty decomposition-separating aleatoric from epistemic sources-is particularly crucial in this setting, yet existing methods, designed for standard generation tasks, fail to capture the unique dynamics of ICL. To address this, we introduce a concept of self-function vectors, built upon Bayesian views and the mechanistic interpretability of ICL. These vectors leverage internal model representations to model the latent concept learned during in-context prompting, thereby enabling a direct estimation of aleatoric uncertainty within a Bayesian framework and circumventing the reliance on brittle input or decoding manipulations. Given the lack of established benchmarks and suitable evaluation protocols, we also propose the first and rigorous evaluation protocol, in which data is manipulated in controlled ways so as to quantify aleatoric uncertainty precisely and separately from epistemic uncertainty. With this new evaluation framework, initially grounded in synthetic tasks for conceptual development and subsequently extended to real-world datasets, we show that our proposed methodology can measure uncertainty of LLM predictions made under ICL more reliably than existing alternative methods. Moreover, we show it can be used as a practical tool for trustworthy-related applications, such as hallucination detection. Our findings pave a new direction for connecting the quantitative view of uncertainty with the mechanistic understanding of model behavior.

Sign languages are expressive visual languages used by Deaf and Hard-of-Hearing (DHH) communities. Despite substantial progress in sign-language recognition, translation, and production, advances remain constrained by fragmented datasets, inconsistent annotations, and limited linguistic coverage. Existing benchmarks often fail to reflect real-world communication needs, and systematic analyses of these limitations remain limited. In this survey, we present a comprehensive index of sign-language datasets, covering 120 resources across 35 sign languages. We analyze key challenges such as modality imbalance, annotation granularity, and signer bias, and outline considerations for future dataset design. We also introduce a 24-field Sign-Language Datasheet and release a public GitHub repository (https://github.com/Ginqwerty/Open-Sign-Language) to support standardized documentation and reproducible evaluation. Overall, our work provides a unified and practical foundation for developing inclusive, robust, and scalable sign-language technologies in real-world applications.

We present Density Field State Space Models (DF-SSM), a framework for compressing SSMs to a 1-bit scaffold with int8 low-rank correction. Applied to Mamba-2 1.3B, we achieve a 278 MB model (9.7x smaller than the 2.7 GB FP16 teacher) that runs at 21.4x faster inference on GPU (batch=1, relative to the mamba-ssm reference implementation) while maintaining downstream task performance within 2-4 percentage points of BitMamba-2, a 1.58-bit model trained from scratch on 150B tokens. The distillation itself requires only 32M tokens and 6 hours on a single A100 GPU, though it presupposes a pretrained FP16 teacher. We develop an optimized inference pipeline combining cuBLAS INT8 tensor cores for the scaffold matmul, custom CUDA kernels for stateful SSM and convolution operations, and an AVX-512 CPU backend for efficient deployment on both GPU and CPU. Beyond compression, we investigate the internal knowledge organization of the resulting model, discovering three distinct processing phases: intent classification (layers 0-3, operating in an abstract space with no vocabulary alignment), knowledge retrieval (layers 25-35, where factual associations localize to a 5-layer window), and output formatting (layers 36-47, where category structure dissolves). Through systematic analysis of 445 factual prompts across 19 categories, we find that early-layer classification is syntactic (driven by template structure) rather than semantic, and that the model exhibits well-organized knowledge representations despite weak factual recall–suggesting that representational structure may precede factual strength.

How do multi-turn reasoning systems fail? The expected answer is logical contradiction, in which the system's maintained state becomes unsatisfiable. We show that the dominant mode is instead satisfiable drift, where the internal state stays consistent while the returned answer silently violates prior commitments. We build DRIFT-Bench (Decomposing Reasoning Into Failure Types), a solver-instrumented benchmark of 816 test problems across three constraint domains, and evaluate four methods on it across four open-weight models (8B-120B parameters). MUS-Repair, which feeds minimal unsatisfiable subsets back to the generator, is strongest in every setting (+1.8 to +15.0 pp over the best non-MUS baseline). But the central finding is what repair leaves behind. After structured feedback, models rarely contradict themselves. They forget. Residual errors are 98-100% satisfiable drift across all settings, while contradiction drops to near zero. Reliable multi-turn systems must separately validate that the returned answer respects the maintained state. Code is available at https://github.com/kaons-research/drift-bench.

We study long-context balanced entropic optimal transport (OT) attention on TPU hardware through a stopped-base, fixed-depth tail-refinement surrogate. After a stopped $T$-step Sinkhorn solve, we unroll a short refinement tail and differentiate that surrogate exactly. For the reported $R=2$ TPU path, the backward pass contains four staircase plan factors. We prove an exact one-reference-tile schedule: the $R=2$ score cotangent is a single reference plan tile times an explicit modifier field built from vector cotangents and dual differences. This yields block-wise cost $O((T+R)LW)$, $O(Ld)$ input storage, and $O(L)$ additional HBM usage for fixed head dimension $d$ and band width $W$ on the balanced fixed-support path. We also formalize the current dustbin\_block path as the same unit-target surrogate on an augmented support, so the adjoint schedule lifts to the single-active-dustbin path used in our TPU runs; this bridge is algebraic and does not claim a general KL-unbalanced or arbitrary-capacity gap model. We provide a local surrogate-bias bound, an a posteriori bias certificate, and a projective contraction certificate for strictly positive active blocks. On synthetic masked problems, the optimized kernel matches exact autodiff of the same centered surrogate to within $10^{-5}$–$10^{-10}$. On TPU v6e-8, a four-configuration Pfam screen completes end-to-end, and a promoted balanced $R=2$ run sustains roughly $8.5$ examples per second through a three-hour budget, reaching step $1437$. Held-out Pfam test shards improve reconstruction from $5.57$ to $2.05$ and sparse CE from $5.53$ to $5.30$ relative to step $0$, with CE logged diagnostically rather than optimized directly; target-barycenter alignment metrics do not materially improve, and a deterministic diagonal reference remains stronger on those metrics.

We introduce a novel task of digital battery passport (DBP) conformance classification and introduce the first public benchmark for the task: BatteryPass-12K, created synthetically from real pilot samples. This is as the EU's battery regulation on DBPs comes into effect soon and there exists no public dataset. We evaluated 22 language models (LMs) in zero-shot inference, spanning small LMs (SLMs), mixture of experts (MoEs), and dense LLMs. We also conducted analysis, additional evaluations of few-shot inference and prompt-injection attacks to find that (1) Thinking models have the best performance (with GPT-5.4 scoring 0.98 (0.03) and 0.71 (0.22) on average as F1 (and confidence interval at 95%) on the validation and test sets, respectively), (2) few-shot examples improve performance significantly, (3) generally capable frontier models find the task challenging, (4) merely scaling model parameters does not necessarily lead to improved performance, as SLMs outperformed some LLMs, and (5) prompt-injection attacks degrade performance. We note that BatteryPass-12K, though limited to real pilot samples, may be useful for other known or emerging tasks in the battery domain, e.g. lifecycle reasoning. We publicly release the dataset under a permissive licence (CC-BY-4.0).

With recent semantic query processing engines, semantic aggregation has become a primitive operator, enabling the reduction of a relation into a natural language aggregate using an LLM. However, the resulting semantic aggregate may contain claims that are not grounded in the underlying relation. Verifying such claims is challenging: they often involve quantifiers, groupings, and comparisons over relations that far exceed LLM context windows and require a costly combination of semantic and symbolic processing. We present Evergreen, a system that recasts claim verification as a semantic query processing task with tailored optimizations and provenance capture. Evergreen compiles each claim into a declarative semantic verification query that can execute on the same query engine used to produce the aggregate. To reduce cost, Evergreen avoids unnecessary LLM calls through verification-aware optimizations, including early stopping, relevance sorting, and estimation with confidence sequences, as well as general-purpose optimizations for semantic queries, such as operator fusion, similarity filtering, and prompt caching. Each verdict is accompanied by citations that identify a minimal set of tuples justifying the result, with semantics based on semiring provenance for first-order logic. On a benchmark of production-inspired workloads over restaurant review and customer support datasets, Evergreen's optimized configurations occupy the entire cost-quality Pareto frontier. With a strong LLM, Evergreen preserves verification quality at an F1 of 0.94 while reducing cost by 3.1x relative to unoptimized verification; with a substantially weaker LLM, it surpasses the strongest external baseline's F1 (0.87 vs. 0.83) at 7.0x lower cost.

The integration of Large Language Models (LLMs) with Retrieval-Augmented Generation (RAG) has significantly advanced Knowledge Graph Question Answering (KGQA). However, existing LLM-driven KGQA systems act as stateless planners, generating retrieval plans in isolation without exploiting historical query patterns: analogous to a database system that optimizes every query from scratch without a plan cache. This fundamental design flaw leads to schema hallucinations and limited retrieval coverage. We propose CacheRAG, a systematic cache-augmented architecture for LLM-based KGQA that transforms stateless planners into continual learners. Unlike traditional database plan caching (which optimizes for frequency), CacheRAG introduces three novel design principles tailored for LLM contexts: (1) Schema-agnostic user interface: A two-stage semantic parsing framework via Intermediate Semantic Representation (ISR) enables non-expert users to interact purely in natural language, while a Backend Adapter grounds the LLM with local schema context to compile executable physical queries safely. (2) Diversity-optimized cache retrieval: A two-layer hierarchical index (Domain $\rightarrow$ Aspect) coupled with Maximal Marginal Relevance (MMR) maximizes structural variety in cached examples, effectively mitigating reasoning homogeneity. (3) Bounded heuristic expansion: Deterministic depth and breadth subgraph operators with strict complexity guarantees significantly enhance retrieval recall without risking unbounded API execution. Extensive experiments on multiple benchmarks demonstrate that CacheRAG significantly outperforms state-of-the-art baselines (e.g., +13.2% accuracy and +17.5% truthfulness on the CRAG dataset).

An effective way to scale up test-time compute of large language models is to sample multiple responses and then select the best one, as in Grok Heavy and Gemini Deep Think. Existing selection methods often rely on external reward models, which requires training a strong reward model and introduces additional computation overhead. As an alternative, previous approaches have explored intrinsic signals, such as confidence and entropy, but these signals are noisy with naive aggregation. In this work, we observe that high-entropy tokens tend to cluster into consecutive groups during inference, providing a more stable notion of model uncertainty than individual tokens. Together, these clusters reveal temporal patterns of model uncertainty throughout the inference process. Motivated by this observation, we propose to use the temporal structure of uncertainty as an intrinsic reward. To this end, we first formalize the basic unit of segment-level uncertainty as the High Entropy Phase (HEP), a variable-length segment that begins at a high-entropy token and ends when consecutive low-entropy tokens appear. We then define the Entropy Centroid, inspired by the concept of the center of mass in physics, as the weighted average position of all HEPs along the trajectory. Intuitively, a lower centroid indicates early exploration followed by confident generation, which we find often corresponds to higher response quality. Based on this insight, we propose the Lowest Centroid method, which selects the response with the lowest entropy centroid among multiple candidates. Experiments on mathematics, code generation, logical reasoning, and agentic tasks, across model scales ranging from 14B to 480B, show that Lowest Centroid consistently outperforms existing baselines and delivers stable gains as model size increases. Code is available at https://github.com/hkust-nlp/entropy-centroid.

Adapting large language models (LLMs) to a targeted task efficiently and effectively remains a fundamental challenge. Such adaptation often requires iteratively improving the model toward a targeted task, yet collecting high-quality human-labeled data to support this process is costly and difficult to scale. As a result, synthetic data generation has emerged as a flexible and scalable alternative. One straightforward approach is through an iterative generation-training loop, where candidate data are synthesized through an external generator, the model is updated using these data and the process is repeated over iterations. However, generated samples can be noisy, highly redundant, or even misaligned with the targeted task distribution. Training indiscriminately on such data can dilute useful learning signals and even degrade model performance. To address this, we introduce a refined paradigm, namely an iterative generation-selection-training loop, which incorporates a selection step prior to model updates. Building on this paradigm, we propose EvoSelect, a data-efficient framework to evolve LLM effectively. Given candidate samples produced by the data generator, EvoSelect selects training data by jointly modeling targeted task alignment and diversity. We estimate task relevance through optimal transport with proxy gradient representations, which quantifies how well candidate samples align with the targeted task distribution. To mitigate redundancy, we incorporate a diversification mechanism that promotes coverage of complementary training samples. By interleaving alignment and diversification, EvoSelect enables progressive LLM evolution toward targeted tasks. Extensive experiments on various benchmarks demonstrate that with either weak or strong data generators, EvoSelect consistently improves adaptation efficacy over existing data selection methods.

Recent work has shown that a model's input word embeddings can serve as effective control variables for steering its behavior toward outputs that satisfy desired properties. However, this has only been demonstrated for pretrained text-completion models on the relatively simple objective of reducing surface-level profanity in short continuations. A natural and practically important question is how well input embeddings can control aligned models, which produce an imbalanced bimodal refuse-or-comply output distribution rather than the smooth distribution characteristic of open-ended generation. We explore this in the context of safety, showing that input word embeddings can be optimized in a sub-lexical manner to minimize the semantic harmfulness of aligned model responses. Our approach uses zeroth-order gradient estimation of a black-box text-moderation API with respect to the input embeddings, and then applies gradient descent on these embeddings to minimize the harmfulness of the generated text. Experiments show that the proposed method can neutralize every safety-flagged response on standard safety benchmarks.

Structural generalization in semantic parsing requires systems to apply learned compositional rules to novel structural combinations. Existing approaches either rely on hand-written algebraic rules (AM-Parser) or fail to generalize structurally (Transformer-based models). We present an alternative requiring no hand-written compositional rules, based on a neural cellular automaton (NCA) with a discrete bottleneck: all compositional rules are learned from data through local iteration. On the SLOG benchmark, the system achieves an overall accuracy of $67.3 \pm 0.2\%$ across 10 seeds (AM-Parser: $70.8 \pm 4.3\%$), with 11 of 17 structural generalization categories at $100\%$ type-exact match, including three where AM-Parser scores $0$–$74\%$. Analysis reveals that all 5,539 failure instances reduce to exactly two mechanisms: novel combinations of wh-extraction context with reduced verb types, and modifiers appearing on the subject side of verbs. When we decompose results by CCG structural features, each sub-pattern either succeeds on all instances or fails on all. Intermediate scores (e.g., $41.4\%$) are mixtures of structurally distinct CCG patterns, not partial generalization. These results suggest that CCG directed types provide higher resolution than SLOG's phenomenon-level categories for characterizing structural generalization, and that the success/failure boundary is determined by the coverage of directed operations in the training data.

AI agents operating on user interfaces must understand how interfaces communicate state and feedback to act reliably. As a core communicative modality, animations are increasingly used in modern interfaces, serving critical functional purposes beyond mere aesthetics. Thus, understanding UI animation is essential for comprehensive interface interpretation. However, recent studies of Vision Language Models (VLMs) for UI understanding have focused primarily on static screenshots, leaving it unclear how well these models handle dynamic UI animations. To address this gap, we created AniMINT, a novel dataset of 300 densely annotated UI animation videos. We systematically evaluate state-of-the-art VLMs on UI animation understanding, including their abilities to perceive the animation effects, identify animation purposes, and interpret animation meaning. Our results show that VLMs can reliably detect primitive motion. However, their high-level animation interpretation remains inconsistent, with substantial gaps relative to human performance. Finally, we use Motion, Context, and Perceptual Cues (MCPC) to probe factors affecting VLM performance, revealing key bottlenecks and directions for future improvement.

Diffusion large language models generate text through iterative denoising, exposing hidden trajectories that may contain reliability signals beyond the final output. We propose HIVE, which compresses trajectory hidden states, selects informative step-layer evidence, and conditions a verifier through continuous prefix embeddings to produce a hallucination score and structured diagnostics. Across two D-LLMs and three QA benchmarks, HIVE outperforms eight established baselines and a verifier-backbone-matched text-only control in all six settings. Relative to text-only verification, hidden-evidence conditioning improves AUROC by 1.73–4.60 points and AUPRC by 1.10–3.62 points, with average gains of 3.15 and 2.28 points, respectively. Ablations, evidence interventions, and cross-dataset transfer further support the complementary value of fine-grained hidden trajectory evidence.

Preserving a speaker's voice identity while generating speech in a different language remains a fundamental challenge in spoken language technology, particularly in specialized domains such as scientific communication. In this paper, we address this challenge through our system submission to the International Conference on Spoken Language Translation (IWSLT 2026), the Cross-Lingual Voice Cloning shared task. First, we evaluate several state-of-the-art voice cloning models for cross-lingual speech generation of scientific texts in Arabic, Chinese, and French. Then, we build voice cloning systems based on the OmniVoice foundation model. We employ data augmentation via multi-model ensemble distillation from the ACL 60/60 corpus. We investigate the effect of using this synthetic data for fine-tuning, demonstrating improvements in intelligibility (WER & CER) and speaker similarity (SIM), with gains varying across languages.

Large language model agents have made strong progress on software engineering, yet current systems suffer from a context coupling problem: the standard code editing interface conflates code inspection, modification planning, and edit execution within a single context window, forcing agents to interleave exploratory viewing with strictly formatted edit generation. Irrelevant context accumulates and edit reliability degrades. We propose SWE-Edit, which decomposes the editing interface into two specialized subagents: a Viewer that extracts task-relevant code on demand, and an Editor that executes modifications from high-level natural language plans – letting the main agent focus on reasoning while delegating context-intensive operations to clean context windows. On SWE-Bench Verified, this decomposition raises resolve rate by 2.1 pp and cuts inference cost by 17.9%, with consistent gains across multiple reasoning-model families (Kimi-K2, MiniMax-M2.1, GLM-4.7). We further show that effective edit-format selection can be trained into a small model rather than requiring frontier-scale capacity: GRPO training on Qwen3-8B with an adaptive find-replace/whole-file-rewrite policy improves edit success by 12.5 pp and brings an 8B open-source editor to parity with GPT-5-nano on downstream SWE-Bench resolve rate. To enable rapid editor iteration, we release PR-Edit, a lightweight evaluation whose scores correlate strongly with SWE-Bench resolve rate. We release our code at https://github.com/microsoft/SWE-Edit.

Safety evaluations of large language models (LLMs) typically report binary outcomes, i.e. attack success rate (ASR), refusal rate, or harmful versus safe classification, which hide how risk changes between prompt and response. We present a paired analysis over human labeled prompt and response records across four harm categories (Sexual, Self harm, Hate and Violence) and ordinal severity levels (Safe, Low, Medium, High). 61% of responses reduce harm relative to the prompt, 36% preserve severity, and 3% escalate. The escalation splits into two mechanisms: benign prompts triggering unrequested harmful detail, and answers that stay on task at higher severity than the prompt. Category decomposition shows that Sexual content exhibits the highest harm persistence in this sample, driven by compliance at the same severity rather than drift from benign inputs. Joint relevance analysis exposes a helpfulness versus harmlessness tradeoff: compliance escalations remain highly relevant, whereas safe responses include generic refusals with low relevance. A public supporting evaluation over 600 prompts and six models reproduces the framework's measurements and two directional signals, while few-shot LLM graders exhibit a prompt/response detection asymmetry that data calibration does not close. Grader prompts, public-evaluation artifacts, and analysis code are shared at https://github.com/microsoft/PairedSafety.

This paper presents a principled and scalable framework for systematically generating complex Question Answering (QA) data. In the core of this framework is a graphlet-anchored generation process, where small subgraphs from a Knowledge Graph (KG) are used in a structured prompt to control the complexity and ensure the factual grounding of questions generated by Large Language Models. The first instantiation of this framework is BioGraphletQA, a new biomedical KGQA dataset of 119,856 QA pairs. Each entry is grounded in a graphlet of up to five nodes from the OREGANO KG, with most of the pairs being enriched with relevant document snippets from PubMed. We start by demonstrating the framework's value and the dataset's quality through evaluation by a domain expert on 106 QA pairs, confirming the high scientific validity and complexity of the generated data. Secondly, we establish its practical utility by showing that augmenting downstream benchmarks with our data improves accuracy on PubMedQA from 49.2% to 68.5% in a low-resource setting, and on MedQA from a 41.4% baseline to 44.8% in a full-resource setting. Our framework provides a robust and generalizable solution for creating critical resources to advance complex QA tasks, including MCQA and KGQA. All resources supporting this work, including the dataset (https://zenodo.org/records/17381119) and framework code (https://github.com/ieeta-pt/BioGraphletQA), are publicly available to facilitate use, reproducibility and extension.

Usability testing with experts and potential users can assess the effectiveness, efficiency, and user satisfaction of graphical user interfaces (GUIs) but doing so remains a costly and time-intensive process. Prior work has used computer use agents (CUAs) and other generative agents that can simulate user interactions and preference, but we show that agents still struggle to provide accurate usability assessments. In this work, we present a novel machine learning method that operationalizes a computational definition of usability to train CUAs to assess GUI usability by i) prioritizing important interaction flows, ii) executing them through human-like interactions, and iii) predicting a learned numerical usability score. We train a computer use agent, uxCUA, with our algorithm on a large-scale dataset of fully interactive user interfaces (UIs) paired with usability labels and human preferences. We show that uxCUA outperforms larger models in accurate usability assessments and produces realistic critiques of both synthetic and real UIs. More broadly, our work aims to build a principled, data-driven foundation for automated usability assessment in HCI.

We present a unified quantitative analysis of the Currier A/B language distinction in the Voynich Manuscript, proceeding in two stages. First, we confirm that the distinction is genuine: a Beta-Binomial mixture model applied to character-pair substitution ratios across 185 folios, without access to Currier's labels, selects 2 by BIC and predicts held-out folio labels at 89% accuracy. Second, we show that the A/B contrast is not primitive but is a low-resolution projection of a higher-dimensional generative system. Its dominant component is a discrete boolean "switch" set once per folio, governing the vowel following the digraphs ch and sh. A two-state binomial mixture achieves Delta\ AIC = 2,549 over a single-state model and assigns 195 of 197 folios unambiguously. This switch does not operate uniformly: word templates divide into fixed contexts and switchable contexts, with template identity accounting for 92% of variance.

Recursive or looped language models have recently emerged as a new scaling axis by iteratively refining the same model computation over latent states to deepen reasoning. We extend such scaling principle from a single model to multi-agent systems, and ask: Can agent collaboration itself be scaled through recursion? To this end, we introduce RecursiveMAS, a recursive multi-agent framework that casts the entire system as a unified latent-space recursive computation. RecursiveMAS connects heterogeneous agents as a collaboration loop through the lightweight RecursiveLink module, enabling in-distribution latent thoughts generation and cross-agent latent state transfer. To optimize our framework, we develop an inner-outer loop learning algorithm for iterative whole-system co-optimization through shared gradient-based credit assignment across recursion rounds. Theoretical analyses of runtime complexity and learning dynamics establish that RecursiveMAS is more efficient than standard text-based MAS and maintains stable gradients during recursive training. Empirically, we instantiate RecursiveMAS under 4 representative agent collaboration patterns and evaluate across 9 benchmarks spanning mathematics, science, medicine, search, and code generation. In comparison with advanced single/multi-agent and recursive computation baselines, RecursiveMAS consistently delivers an average accuracy improvement of 8.3%, together with 1.2$\times$-2.4$\times$ end-to-end inference speedup, and 34.6%-75.6% token usage reduction. Code and Data are provided in https://recursivemas.github.io.

Real-world data visualization (DV) requires native environmental grounding, cross-platform evolution, and proactive intent alignment. Yet, existing benchmarks often suffer from code-sandbox confinement, single-language creation-only tasks, and assumption of perfect intent. To bridge these gaps, we introduce DV-World, a benchmark of 260 tasks designed to evaluate DV agents across real-world professional lifecycles. DV-World spans three domains: DV-Sheet for native spreadsheet manipulation including chart and dashboard creation as well as diagnostic repair; DV-Evolution for adapting and restructuring reference visual artifacts to fit new data across diverse programming paradigms and DV-Interact for proactive intent alignment with a user simulator that mimics real-world ambiguous requirements. Our hybrid evaluation framework integrates Table-value Alignment for numerical precision and MLLM-as-a-Judge with rubrics for semantic-visual assessment. Experiments reveal that state-of-the-art models achieve less than 50% overall performance, exposing critical deficits in handling the complex challenges of real-world data visualization. DV-World provides a realistic testbed to steer development toward the versatile expertise required in enterprise workflows. Our data and code are available at \href{https://github.com/DA-Open/DV-World}{this project page}.

How much does a user's skill with AI shape what AI actually delivers for them? This question is critical for users, AI product builders, and society at large, but it remains underexplored. Using a richly annotated sample of 27K transcripts from WildChat-4.8M, we show that fluent users take on more complex tasks than novices and adopt a fundamentally different interactional mode: they iterate collaboratively with the AI, refining goals and critically assessing outputs, whereas novices take a passive stance. These differences lead to a paradox of AI fluency: fluent users experience more failures than novices – but their failures tend to be visible (a direct consequence of their engagement), they are more likely to lead to partial recovery, and they occur alongside greater success on complex tasks. Novices, by contrast, more often experience invisible failures: conversations that appear to end successfully but in fact miss the mark. Taken together, these results reframe what success with AI depends on. Individuals should adopt a stance of active engagement rather than passive acceptance. AI product builders should recognize that they are designing not just model behavior but user behavior; encouraging deep engagement, rather than friction-free experiences, will lead to more success overall. Our code and data are available at https://github.com/bigspinai/bigspin-fluency-outcomes