Papers

When distilling reasoning from large language models (LLMs) into smaller ones, teacher rationales for similar problems often vary wildly in structure and strategy. Like a chef who makes the same dish differently each time, this inconsistency burdens the student with noisy supervision that is hard to internalize. We propose Distillation through Reasoning Path Compression (D-RPC), which constrains the teacher to follow a compact, dynamically maintained bank of reusable high-level reasoning paths. For each training question, D-RPC retrieves the most relevant path and conditions the teacher to follow it, producing rationales that are consistent across similar problems yet diverse enough to cover different problem types. A PAC-Bayes analysis formalizes a trade-off between bank size and coverage, with smaller banks limiting path diversity at the risk of coverage gaps. Our ablations favor an intermediate bank size, consistent with this trade-off. Across five math and commonsense reasoning benchmarks with two student models, D-RPC achieves the highest average accuracy among the evaluated distillation methods, while using fewer tokens than template-heavy alternatives.

Web agents perceive web pages through an observation space, yet its granularity has remained an underexamined design choice. Existing work treats observation at the same element-level granularity as the action space, leaving the page's functional organization implicit and forcing the agent to infer it from element-level signals at every step. We argue observation should instead operate at the granularity of functional regions, parts of the page that each serve a distinct purpose. We propose Region4Web, a framework that reorganizes the AXTree into functional regions through hierarchical decomposition and semantic abstraction, exposing the page's functional organization as the basis for page state understanding. Moreover, we propose PageDigest, a web-specific inference pipeline that delivers this region-level observation to the actor agent as a compact per-page digest that persists across steps. On the WebArena benchmark, PageDigest substantially reduces observation length while improving overall task success rate across diverse backbone large language models (LLMs) and established agent methods, regardless of backbone capacity. These results show that operating at the granularity of functional regions delivers a more compact and informative basis for the actor agent than element-level processing alone.

Modern large language models (LLMs) can find a needle in a haystack (locating a single relevant fact buried among hundreds of thousands of irrelevant tokens) with near-saturated accuracy, yet fail to retrieve the last few items in a short list. We call this failure the Position Curse. For instance, even in a two-line code snippet, Claude Opus 4.6 misidentifies the second-to-last line most of the time. To characterize this failure, we evaluated two complementary queries: given a position in a sequence (of letters or words), retrieve the corresponding item; and given an item, return its position. Each position is specified as a forward or backward offset from an anchor, either an endpoint of the list (its start or end) or another item in the list. Across both open-source and frontier closed-source models, backward retrieval substantially lags forward retrieval. To test whether this capability can be rescued by post-training, we constructed PosBench, a position-focused training dataset. LoRA fine-tuning improves both forward and backward retrieval and generalizes to a held-out code-understanding benchmark (PyIndex), yet absolute performance remains far from saturated. As LLM coding agents increasingly operate over large codebases where precise indexing becomes essential for code understanding and editing, position-based retrieval emerges as a key capability for future pretraining objectives and model design.

Recent literature on fine-tuning Large Language Models highlights a fundamental debate. While Full Fine-Tuning (FFT) provides greater representational plasticity, Low-Rank Adaptation (LoRA) can match or surpass FFT performance while constraining updates to a low-rank space and potentially benefiting from additional regularization. Through empirical evaluation across diverse tasks (SQL, Medical QA, and Counterfactual Knowledge) and varying language models (Gemma-3-1B, Qwen2.5-1.5B, and Qwen2.5-3B), we observe both trends and find that the better static architecture depends on the task and model. Spectral and truncation analyses further show that endpoint compressibility alone does not explain these task differences, suggesting task-score sensitivity and constrained optimization trajectories as possible explanations. To address this challenge, we propose a Mixture of LoRA and Full (MoLF) Fine-Tuning, a unified framework that enables continuous navigation between both training regimes. MoLF dynamically routes updates between FFT and LoRA at the optimizer level to ensure that exact gradient signals are available to both experts throughout training, while only selected experts update their weights. For memory-constrained environments, we also introduce MoLF-Efficient, which freezes base weights and only routes updates among a pair of LoRA experts of potentially varying rank. Our evaluations show that MoLF either improves on or stays within $1.5$ percentage points of the better of FFT and LoRA across the nine tested settings, while MoLF-Efficient outperforms both AdaLoRA and AdaMix in eight of nine settings, with gains over the stronger baseline of up to $11.70$ percentage points on Fact, $3.13$ on Med, and $2.98$ on SQL.

Computer-use agents(CUAs)are moving frombounded benchmarks toward real software environments, wherethey operate browsers, desktops, mobile applications, flesystems,terminals, and tool backends. In such settings, reliability isno longer captured by task success alone: perception errors,planning drift, memory use, tool mediation, permission scope,and runtime oversight jointly determine whether agent actionsremain aligned with user intent, Existing surveys organize theCUA landscape by methods, platforms, benchmarks, or securitythreats, but less explicitly connect capability formation, author-ity exposure, failure manifestation, and control placement. Toaddress this gap, the article develops an architecture-lifecycleframework for deployment-grounded reliability in CUAs. Thearchitectural view analyzes Perception, Decision, and Executionas coupled layers that transform software observations intoauthority-bearing actions, The lifecycle view examines Creation.Deployment, Operation, and Maintenance as stages in which priorsare learned, tools and permissions are bound, runtime trajecto.ries are stressed, and assurance must be preserved under drift.Using this lens, the analysis synthesizes representative systems,benchmarks, and security/privacy studies; distinguishes wherefailures become visible from where their enabling conditions areintroduced, and maps recurring intervention surfaces for controloversight, and assurance. OpenClaw is used only as a public moti.vating example of an open deployment pattern, not as a verifedinternal case study. The conclusion highlights open challengesin controllable grounding, long-horizon constraint preservation,safe authority binding, mixed-trust runtime defense, privacy-preserving memory,and continual assurance.

Multimodal Large Language Models (MLLMs) have made remarkable progress on vision-language reasoning, yet most methods still compress visual evidence into discrete textual thoughts, creating an information bottleneck for fine-grained perception. Recent latent visual reasoning methods attempt to reason in continuous hidden states, but we find that they suffer from insufficient manifold compatibility: latent trajectories drift away from pretrained reasoning circuits, collapse into instance-agnostic patterns, and are often bypassed during answer generation. To address these issues, we propose RIS (Retrieve, Integrate, and Synthesize), a spatial-semantic grounded framework that develops latent reasoning as a compatible extension of pretrained MLLM computation. We first construct a step-wise grounded reasoning dataset with bounding boxes and region-specific semantic descriptions. Built on this supervision, RIS anchors latent tokens to both spatial and semantic evidence, enforces their causal role through a progressive attention bottleneck, and introduces short language transition tokens to bridge synthesized latent states back to vocabulary-aligned decoding. Experiments on V*, HRBench4K, HRBench8K, MMVP, and BLINK show consistent improvements over closed/open-source and latent reasoning baselines. Further analyses demonstrate that RIS learns diverse, interpretable, and progressively integrated latent trajectories, offering a practical path toward faithful internal visual reasoning in MLLMs.

Language model (LM) alignment improves model outputs to reflect human preferences while preserving the capabilities of the base model. The most common alignment approaches are (i) reinforcement learning, which maximizes the expected reward under a KL-divergence constraint, and (ii) best-of-$N$ alignment, which selects the highest-reward output among $N$ independent samples. Despite their widespread use, the fundamental limits of reward improvement under a KL budget remain poorly understood. We characterize the information-theoretic limits of KL-regularized alignment by deriving the maximum achievable expected reward gain for a fixed KL-divergence budget. Our first result provides a closed-form expression for the optimal reward improvement, governed by a Jeffreys divergence term rather than the $\sqrt{\texttt{KL}}$ used in prior analyses. We further reformulate this expression as a covariance under the base model, yielding a practical estimator that predicts achievable alignment gains from base model samples alone. We extend our analysis to the proxy reward setting, showing that the gap between ideal and proxy alignment (reward hacking) grows with the magnitude of reward error and when the KL penalty factor decreases. We then prove that reward ensembling mitigates reward hacking, providing a theoretical justification for this technique used in practice. Empirically, we compute the KL-reward Pareto frontier for two tasks for LMs, safety and summarization, and show that best-of-$N$ closely approaches the theoretical limit, while PPO and GRPO remain substantially suboptimal. Our theoretical results shed light on several empirically observed phenomena in the alignment literature and suggest that algorithmic improvements are needed to achieve optimal alignment without high inference costs.

Evaluating literary quality requires assessing interpretive dimensions such as cultural representation, emotional depth, and philosophical sophistication that resist straightforward computational measurement. We introduce SAGE, a hierarchical evaluation framework that decomposes literary quality into ontology-grounded interpretive dimensions assessed through structured large language model evaluation with multi-round iterative reflection and independent validation. We validate the framework on 100 short stories (50 canonical works, 30 pulp fiction, 20 LLM-generated narratives) across three analytical layers (cultural, emotional-psychological, existential-philosophical) using dual-mode assessment. Across 600 evaluations, the framework achieves 98.8% score convergence and greater than 94% inter-rater agreement, with near-perfect mode invariance between content-based and metadata-based evaluation. Statistical analysis reveals a consistent genre hierarchy (Canonical > Pulp > LLM, all p<0.001) with layer-specific discrimination: cultural critique and philosophical depth exhibit very large effect sizes (Cohen's d>2.4), while emotional representation shows smaller gaps (d=1.68), suggesting that affective patterns are more learnable from training data than critical stance or philosophical depth. Cross-layer correlations (r=0.649-0.683) confirm the three dimensions capture empirically distinguishable quality facets. These findings demonstrate that theory-driven LLM evaluation can achieve measurement-grade reliability and support systematic identification of where current generative models fall short of human literary production, with direct implications for scalable automated evaluation of open-ended text generation.

The Translation Tax is often treated as a scalar: translated benchmarks are assumed to inflate scores by preserving English-source cues. We audit this claim in an English-to-Chinese setting. Three proxy estimators disagree: back-translation gaps are small and parser-fragile; cue-score calibration does not predict item-level gains; and a six-model native-control comparison shows model-family rather than uniform benchmark effects. We add a same-item LLM-naturalization stress test that holds answer, options, and content fixed while rewriting Chinese surface form. After correcting a prompt-construction bug, this contrast no longer supports a model-family interaction, but it preserves a residue dose-response: high-residue items benefit while low-residue items do not. The result is not a single Translation Tax, but a set of estimator- and item-dependent validity risks. We release per-cell evidence, the naturalization protocol, human QC, and a reporting checklist for translated multilingual benchmark papers.

Automatic speech recognition (ASR) evaluation compares system output to ground truth transcripts, with Word Error Rate (WER) quantifying the distance between them. But ground truth transcripts are not discovered - they are produced by human annotators following conventions that encode normative assumptions about which speech features matter. Different conventions (verbatim, non-verbatim, legal) produce different transcripts of identical speech and judge the same ASR output differently. This paper argues that reference monism - enforcing a single transcription convention as ground truth - commits epistemic injustice. Speakers with aphasia, whose speech includes clinically meaningful disfluencies, are systematically disadvantaged when evaluated against "clean" references that treat those disfluencies as errors. The harm is not merely differential performance, but that evaluative infrastructure lacks interpretive resources to recognize their contributions as legitimate. We develop a philosophical framework introducing the hermeneutical gap, formalize Epistemic Injustice Distance (EID) to measure reference monism's cost, and demonstrate empirically using AphasiaBank that WER varies depending on which convention defines ground truth. We propose WER-Range: reporting performance across legitimate conventions rather than assuming a single correct answer.

Large language models (LLMs) increasingly receive information as streams of passages, conversations, and long-context workflows. While longer context windows expose more evidence, they do not ensure that useful information is preserved and reused. We study continual context consolidation: writing current context into model weights while limiting interference with previously consolidated information. We propose Self-Consolidating Language Models (SCoL), a post-training framework in which, given current context, an LLM learns to generate textual update instructions specifying which of its own Transformer layers should be updated. Because committed updates change the model that later generates future selections, we train SCoL with meta-reinforcement learning over an evolving model state. We instantiate SCoL with supervised QA rewards on SQuAD knowledge incorporation and intrinsic likelihood-based rewards for LongBench v2 long-context consolidation. Across both settings, SCoL improves acquisition and retention over prompting, summarization, batch test-time training, and sequential finetuning baselines. Analysis of learned selection patterns shows that SCoL encourages the LLM to generate sparse update locations that align with layers of high Fisher information, suggesting that the model learns to route plasticity toward loss-sensitive regions while limiting interference. Moreover, SCoL transfers from shorter meta-training streams to longer LongBench v2 streams at evaluation, suggesting that our framework supports scalable streaming consolidation.

The LLM Wiki pattern, to compile and provide domain knowledge into a persistent artifact and serve it to LLMs via KV cache inference, promises context access at sub-second latency with zero retrieval failure. Realizing this requires solving the compilation gap: LLM compilation distilling raw documents into a wiki without catastrophically discarding critical facts. We characterize this gap across 17 RepLiQA domains (6,800 questions): we observe that full context KV cache inference outperforms RAG on curated knowledge (4.38 vs. 4.08 out of 5, 7.3 faster TTFT) but degrades below RAG at scale due to attention dilution, and blind compilation fails entirely (2.14 to 2.32 vs. 3.46, 53 to 60% catastrophic failure rate). To address the compilation gap, we propose WiCER (Wiki-memory Compile, Evaluate, Refine), an iterative algorithm inspired by counterexample-guided abstraction refinement (CEGAR) that closes this gap. WiCER evaluates compiled wikis against diagnostic probes, identifies dropped facts, and forces their preservation in subsequent compilations. One to two iterations recover 80% of lost quality (mean 3.24 vs. 3.47 for raw full-context across the 15 topics with baselines), reducing catastrophic failures by 55% relative. An ablation across all 17 topics confirms that targeted diagnosis (+0.95), not generic pinning (+0.16), drives the gains. All code and benchmarks are released for reproducible research.

Real-world clinical diagnosis is a complex process in which the doctor is required to obtain information from both interaction with the patient and conducting medical exams. Additionally, the doctor needs to adapt to different patient personas, as well as noisy and incomplete information that can happen at any time during the process. However, existing benchmarks for medical LLMs and methods for automatic diagnosis largely simplify this process by reducing it to single-turn question answering, noise-free conversations, or sequential exam making, etc., ignoring the interactive and uncertain nature of clinical diagnosis. In this paper, we aim to address this gap by formalizing clinical diagnosis as a Partially Observable Markov Decision Process (POMDP) with three action types: questioning the patient, ordering medical exams as tool calls, and issuing a diagnosis. We also introduce a systematic noise model comprising seven patient noise types and three exam noise types. Using our proposed environment, we train an effective diagnosis agent, MedExAgent, through a two-stage pipeline that first performs supervised finetuning on synthetic conversations structured after the Calgary-Cambridge model for clinical interviews, and then applies DAPO to optimize a composite reward capturing diagnostic accuracy, tool call quality, and exam cost including financial cost and patient discomfort. Through extensive experiments and ablation studies, we demonstrate that MedExAgent achieves diagnostic performance comparable to larger models while maintaining cost-efficient examination strategies.

Benchmarks like GSM8K are popular measures of mathematical reasoning, but leaderboard gains can overstate true capability due to memorization of fixed test sets. Most robustness variants apply surface-level perturbations (paraphrases, renamings, number swaps, distractors) that largely preserve the underlying facts, and static releases can themselves become memorization targets over time. We introduce GSM-SEM, a reusable and stochastic framework for generating semantically diverse benchmark variants with substantially higher semantic variance than prior approaches. GSM-SEM perturbs problem statements by modifying entities, attributes, and/or relationships, frequently altering underlying facts and requiring models to recompute solutions under new conditions, while constraining generation to preserve the original calculations/answer and approximate problem difficulty. GSM-SEM generates fresh variants on each run without requiring re-annotation, reducing reliance on static public benchmarks for evaluation and thereby lowering the bias of memorization. We apply GSM-SEM on GSM8K and two existing variation suites (GSM-Symbolic and GSM-Plus), producing GSM8K-SEM, GSM-Symbolic-SEM, and GSM-Plus-SEM. Evaluating 14 SOTA LLMs, we observe consistent performance drops with larger decline when semantic perturbations are coupled with symbolic/plus variations (average drop rate 28% in maximum strictness configuration of GSM-SEM). We publicly release the three SEM variants as fully human-validated datasets. Finally, to demonstrate applicability beyond GSM-style math problems, we apply GSM-SEM to additional benchmarks including BigBenchHard, LogicBench, and NLR-BIRD.

Large Language Models (LLMs) have shown good performance on various science educational benchmarks, demonstrating their potential for use in science and mathematics education. Yet, LLMs tend to be evaluated on science and mathematical educational datasets from the Western world, with an underrepresentation of datasets from the Global South. Furthermore, they tend to have multiple-choice answer options that are trivial to evaluate. In this work, we present NSMQ Riddles, a novel benchmark of Scientific and Mathematical Riddles from Ghana's National Science and Maths Quiz (NSMQ) competition to evaluate LLMs. The NSMQ is an annual live TV competition for senior secondary school students in Ghana that brings together the smartest high school students in Ghana who compete in teams of 2 by answering questions in biology, chemistry, physics, and math over five rounds and five stages until a winning team is crowned for that year. NSMQ Riddles consists of 11 years of riddle questions (n=1.8K) from the 5th round, with each riddle containing a minimum of 3 clues. Students compete to be the first to guess the answer on any of the clues, with earlier clues being vague and also fetching more points. The answers are usually a number, word, or short phrase, allowing for automatic evaluation. We evaluated state-of-the-art models: closed (GPT-5.4, Gemini 3.1 Pro, Claude Opus 4.6) and open models (Kimi-K2.5, DeepSeek-V3.1, GPT-OSS-120B) with high and low reasoning settings. Our evaluation shows that the dataset is challenging even for state-of-the-art LLMs, which performed worse than the best student contestants. This work contributes a novel and challenging benchmark for scientific and mathematical reasoning from the Global South towards enabling a true global benchmarking of LLMs' capabilities for science and mathematics education.

Human dialogue involves more than exchanging information; it also expresses beliefs, emotions, and subjective cognitive styles. Yet current AI dialogue systems often enforce semantic uniformity, sacrificing diversity and interpretability. We present MAPS (Multi-Agent Perspective Spaces), a novel framework that models dialogue between cognitively distinct agents through domain-weighted profiles, dynamic GRU-based memory, and interpretable token-level attention. MAPS enables agents to maintain individualized reasoning while progressively converging on shared meaning. Evaluations on EmpatheticDialogues, TopicalChat, and MultiWOZ show that MAPS supports semantic alignment without collapsing subjectivity. Our results demonstrate a path toward cognitively grounded, interpretable dialogue systems that balance expressiveness and coherence.

Probing the capabilities of Large Language Models (LLMs) and building robust solutions for Multiple-Choice Question Answering (MCQA) remain central challenges in natural language understanding. Furthermore, the rapid proliferation of LLMs has created the implicit assumption that more sophisticated prompting techniques yield better performance. Several studies claim such gains, but report them under differing models, prompt wordings and answer-extraction rules, so the gains cannot be attributed to the technique alone. We address this gap with ReasonLab, an evaluation framework in which the prompting technique is a first-class experimental variable alongside the model and the dataset, and which retains every generation for inspection. Using ReasonLab we conduct a controlled study of 8 prompting techniques across 10 MCQA datasets, 27 model configurations and 480,927 evaluations at temperature 0. We find that the prompting technique is a minor determinant of accuracy: on configurations without a reasoning budget the reasoning triggers improve on direct prompting by only 3.92 to 4.69 pp and are indistinguishable from one another, and on configurations with reasoning enabled no technique differs by more than 0.51 pp. Self-Generate is the only technique with a consistent effect, a reduction of 2.95 pp. We further investigate three phenomena: (1) the comparison of models on a common set of datasets, where model size does not predict accuracy, (2) the trade-offs across thinking budgets, where enabling reasoning is worth up to 12.74 pp whereas an eightfold budget increase adds only 0.48 to 2.10 pp, and (3) the variation in dataset difficulty, with 60% of benchmarks below 70% accuracy and a 43.9 pp spread from easiest to hardest. These results suggest that, for MCQA, the prompting technique is a minor lever compared with enabling model reasoning, and that substantial headroom remains.

This paper introduces tool efficiency, a new quantitative metric to evaluate the rate of useful tool calls in an LLM agent trajectory. To ensure that tool efficiency is well-defined, we also introduce marginal tool utility, a new quantitative metric defined per tool call indicating whether a tool is useful or whether it can be safely removed from the tool suite without affecting accuracy while increasing tool efficiency; in this paper, we determine the sign of marginal tool utility for each tool call in a trajectory using LLM-as-a-Judge. While much prior work has been done to develop techniques that improve tool use by LLMs and design evaluation methods measuring efficiency indirectly using accuracy as a proxy, our work is centered on measuring efficiency directly via the quantitative metric proposed in this paper in post hoc trajectory analyses. It is our intention that this work contributes to the frontier of LLM evaluation research as a springboard for future benchmark designs and agent harness engineering (specifically with regards to creating lean tool suites) that optimize for metrics that complement but are distinct from accuracy.

The inference efficiency of diffusion large language models (dLLMs) is constrained by two challenges: bidirectional attention precludes efficient KV-cache reuse, while increasing decoding parallelism with static confidence thresholds can compromise generation quality. We observe that both challenges arise from a shared phenomenon: as tokens are decoded, their contextual integration through bidirectional attention causes token representations to drift (evolve) across decoding steps. This insight motivates Polestar, a training-free inference framework that uses token representation drift as a unified signal to jointly address both challenges. Polestar comprises two components: Polestar-Cache, which identifies stale KV-cache positions via drift and performs sparse KV-cache refreshes to enable efficient reuse, and Polestar-Commit, which detects sharp drift events to reliably identify commit-ready tokens. Across mathematics and coding benchmarks on several dLLM families, Polestar sets a new state of the art on the accuracy-throughput Pareto frontier, achieving up to 10.73% accuracy improvement, up to 3.7x higher throughput, and high decoding parallelism of 3.67 tokens per forward pass over existing baselines.

In this paper we introduce token time continuous diffusion (TTCD), a new diffusion language model which (a) operates in continuous space, deterministically mapping Gaussian noise to a final token canvas with no further sampling, and crucially (b) incorporates a new notion of per-token times, with some tokens proceeding from noise to token at a faster rate than others. Continuous space modeling helps TTCD avoid the parallel sampling of multiple tokens, which is a key source of inaccuracy at high speedups for models that iterate purely in discrete space. The notion of per-token times helps TTCD to better model conditional generation, allows for more sure tokens to proceed at a faster rate, and allows for differentiated inter-token influences during refinement. TTCD outperforms discrete models at high speedups. We train a 160M parameter TTCD model on OpenWebText, and then self-distill it; we find that at high speedups we are comparable in unconditional generation quality, and outperform in conditional generation, several existing models of similar size trained, on the same data, and self-distilled. We achieve similar gains in Sudoku solving as well.

For Large Language Models to reliably answer user queries, users must clearly specify requirements, context, and constraints. In practice, however, user queries are often underspecified, forcing models to infer unstated assumptions that may misalign with the actual user intent. Existing prompt engineering guidelines aim to mitigate this issue, they are typically generic and task-agnostic, limiting their practical utility. Additionally, existing guidelines are formed manually and in a non-systematic way. To this end, we study prompt guideline optimization: the problem of automatically generating task-specific guidelines that help write better-specified prompts for a given task and model. Our key observation is that existing (completed) task examples (aka reference answers) often implicitly encode the missing information required to complete underspecified queries, including behavioral constraints, contextual assumptions, and evaluation criteria. We therefore propose AGOPS, an automatic approach that evolves task-specific guidelines via an optimization scheme that involves a prompt LLM writer, a solver LLM and prompt evolution, which maximize downstream effectiveness on a set of examples (user queries with reference answers). At inference time, our guidelines help users write well-specified prompts, boosting the effectiveness of LLMs. We show across mathematical reasoning, medical question answering, and coding tasks, that prompt underspecification leads to major drops (up to 95.3%) in downstream task performance (compared to well-specified prompts) and, perhaps more importantly, that this drop can hardly be recovered by existing prompt optimization techniques. Users following AGOPS guidelines can regain this loss (increasing performance between 15.5 to 81.7% on average) consistently across all benchmarks.

Large Language Models (LLMs) perform strongly on many language tasks, but their capability in structurally constrained, accessibility-critical modalities such as Braille remains unclear. We evaluate state-of-the-art LLMs on bidirectional Korean-Braille translation using a human-annotated dataset. Despite expectations that multilingual, instruction-tuned models can generalize to Braille via text representations, we find consistently poor, unstable outputs and substantial disagreement with human judgments. These results point to missing Braille-aware tokenization and weak alignment between Korean and Braille patterns. In contrast, supervised fine-tuning of a small model (T5-small) on the same data yields large and stable gains over zero-shot and prompted LLM baselines across standard metrics (SacreBLEU, ChrF++, CER, BLEU, ROUGE-L, METEOR, CIDEr). Our findings reveal a systematic limitation of current LLMs and demonstrate the effectiveness of modest task-specific supervision.

Human-factor event diagnosis is essential for learning from operational events in nuclear power plants, yet its quality depends strongly on expert interpretation of narrative reports and guideline-based reasoning.Existing data-driven or one-shot large language model approaches often lack structured reasoning, have limited alignment with formal diagnostic guidelines, and may generate logically inconsistent conclusions. To address this issue, this study proposes G-SHARE, a guideline-based structured reasoning framework that operationalizes the CNNP nine-step human-factor event diagnosis guideline into a multi-stage diagnostic pipeline.The framework consists of evidence extraction, stepwise diagnostic reasoning, and post-hoc consistency repair, enabling explicit use of report evidence, intermediate rationale generation, and logical validation of diagnostic outputs. A dataset of real human-factor event reports was constructed from Chinese nuclear industry sources, and a gold-standard subset annotated by domain experts was used for evaluation. Results show that G-SHARE substantially outperforms one-shot prompting and traditional machine learning baselines, with the strongest version achieving the best overall accuracy and macro-F1. Ablation results further indicate that structured reasoning and consistency enforcement are critical to robust diagnosis, especially under weak prompting conditions. The findings demonstrate the value of transforming expert diagnostic guidelines into auditable reasoning workflows, providing a practical pathway for intelligent human-factor analysis in safety-critical industries.

Chain-of-Thought (CoT) reasoning improves the performance of Large Language Models (LLMs), yet it writes steps as text and follows one reasoning path at a time, which limits the capabilities of such models. Latent reasoning methods replace these steps with continuous thought representations and claim that a thought representation substitutes for the written reasoning, compresses trajectories, distinguishes semantic content, and achieves superposition of multiple exploration paths at once. These claims concern the content of the thought representation, whereas the evaluations of these methods rely on downstream task accuracy, token counts, and analyses specific to each method. We formalize these claims as a valid thought representation space and prove that a space is valid if and only if its thought representations attain four properties, namely causality, minimality, separability, and stability, each of which follows from one claim. Translating these properties into four measures and auditing three training-free and three trained methods shows gaps in these claims that accuracy may not necessarily reveal. Training-free methods fail all four properties and do not encode the reasoning they are meant to replace, whereas trained methods generally improve substitution. The four properties and measures therefore provide a theoretically grounded evaluation beyond downstream task accuracy and expose failure modes that may guide future development of latent reasoning methods.

We present Grokers, an architecture for building persistent, structured comprehension of typed knowledge graphs through bottom-up inductive traversal of dependency subgraphs. Unlike retrieval-augmented generation (RAG), which pays full comprehension cost at every query, Grokers pushes intelligence to write time: autonomous Groker agents analyze nodes in a typed stream graph, extract structured attributes via governed language model (LM) calls, and inductively compose that understanding upward through dependency relations, writing enriched typed attributes that serve all future queries at zero additional LM cost. We prove three formal properties: (1) the Byte-Identity Theorem, establishing that context blocks assembled from a transactionally-maintained denormalization index are byte-identical across LM turns between semantic changes, enabling KV-cache hit rates approaching 100%; (2) the Accumulation Monotonicity Theorem, establishing that the fraction of interactions resolved without LM calls is non-decreasing in the number of completed interactions under a governed wisdom library growth protocol; and (3) the Dual-Traversal Ordering Theorem, establishing that top-down generation and bottom-up comprehension are the unique correct traversal orderings for their respective tasks over a dependency DAG, and that their composition closes into a complete generation-comprehension cycle. We further present a deterministic alternative to embedding-based semantic search, with a synonym caching protocol whose LM fallback rate converges to zero for finite-vocabulary domains. A reference implementation is provided in the open-source Qbix / Safebox / Safebots stack.

The large-scale digitization of historical archives has created a paradox: "dark data"-digital objects lacking metadata for retrieval. Manual archival description is slow and expensive, limiting discovery and reuse. We propose Vidya, a modular pipeline that orchestrates Large Language Models (LLMs) and FOSS tools to automate semantic enrichment and archival ingestion at scale. Vidya constrains generations using YAML-defined ontologies and Pydantic validation, producing deterministic, structured JSON outputs from probabilistic models. Developed at Laboratory for Digital Humanities and Innovation (LAMUHDI) of the State University of Ponta Grossa (UEPG), Vidya applies Maker principles and open-source practices to enable low-cost deployment in memory institutions using modest hardware. We compare LLM performance and present a cost-benefit analysis showing major gains, reducing processing time from decades to days while complying with NOBRADE and ISAD(G).

Large language models (LLMs) are widely used for tutoring, feedback generation, and content creation, but their broad pretraining makes them hard to constrain and poor substitutes for controllable learners. Educational systems often require inspectable and editable knowledge states: educators want to know what a system assumes the learner knows, and learners benefit when the system can justify actions in terms of explicit skills, misconceptions, and strategies. Inspired by cognitive architectures, we propose Cognitive Agent Compilation (CAC), a framework that uses a strong teacher LLM to compile problem-solving knowledge into an explicit target agent. CAC separates (i) knowledge representation, (ii) problem-solving policy, and (iii) verification and update rules, with the goal of making bounded problem solving more inspectable and editable in educational settings. We present an early proof of concept implemented with Small Language Models that surfaces key design trade-offs, particularly between explicit control and scalable generalization, and positions CAC as an initial step toward bounded-knowledge AI for educational applications.

Diffusion language models (DLMs) promise parallel, order-agnostic generation, but on standard benchmarks they have historically lagged behind autoregressive models in sample quality and diversity. Recent continuous flow and diffusion approaches have narrowed this gap. In this work, we further close the autoregressive gap by modeling text as a continuous diffusion process over fixed-width binary bitstreams. We refer to the resulting model as CoBit (Continuous Bitstream Diffusion). Our approach represents semantic tokens as analog bit sequences and uses a matched-filter residual parameterization to isolate contextual learning from analytic independent-bit posteriors. Crucially, we adopt a stochastic sampler that applies Langevin-type corrections gated by the entropy-rate profile, concentrating stochasticity in high-information regions while remaining nearly deterministic elsewhere. On LM1B, our 130M-parameter model reaches a generative perplexity (GenPPL) of 59.76 at matched real-data entropy (4.31) using 256 neural function evaluations (NFEs), outperforming prior DLM baselines and reaching the autoregressive reference. On OpenWebText (OWT), our sampler establishes a new continuous-DLM Pareto frontier, achieving GenPPL 27.06 at entropy 5.26 using 4x fewer steps than previous 1024-NFE baselines. Scaling the same recipe to a 462M-parameter model (CoBit-M) further improves the OWT GenPPL-entropy frontier over the 130M model (CoBit-S) and over medium-scale continuous and discrete DLM baselines, reaching GenPPL 19.5 at entropy 5.40, near real-data entropy (5.44), and approaching pretrained GPT-2 Medium over the high-quality region. As an additional benefit, bitstream diffusion removes the O(V) vocabulary scaling bottleneck of standard DLMs: by predicting O(log V) bitwise logits via semantic bit-patching, it lowers memory and raises throughput, a scalable paradigm as vocabulary sizes grow.

Architectural code smells erode software maintainability and are costly to repair manually, yet unlike localized bugs, they require cross-module reasoning about design intent that challenges both developers and automated tools. While large language model agents excel at bug fixing and code-level refactoring, their ability to repair architectural code smells remains unexplored. We present the first empirical evaluation of LLM agents on architectural code smell repair. We contribute SmellBench, a task orchestration framework that incorporates smell-type-specific optimized prompts and supports iterative multi-step execution, together with a scoring methodology that separately evaluates repair effectiveness, false positive identification, and net codebase impact. We evaluate 11 agent configurations from four model families (GPT, Claude, Gemini, Mistral) on 65 hard-severity architectural smells detected by PyExamine in the Python project scikit-learn, validated against expert judgments. Expert validation reveals that 63.1% of detected smells are false positives, while the best agent achieves a 47.7% resolution rate. Agents identify false positives with up to $\kappa = 0.94$ expert agreement, but repair aggressiveness and net codebase quality are inversely related: the most aggressive agent introduces 140 new smells. These findings expose a gap between current LLM capabilities in localized code transformations and the architectural understanding needed for cross-module refactoring. SmellBench provides reusable infrastructure for tracking progress on this underexplored dimension of automated software engineering. We release our code and data at https://doi.org/10.5281/zenodo.19247588.

Personalized systems rely on user representations to connect behavioral history with downstream recommendation applications. Existing methods typically employ either supervised latent user embeddings, which are effective for retrieval but difficult to interpret, or textual user profiles, which are interpretable but challenging to optimize for downstream utility due to lack of direct supervision. To bridge this gap, we present BLUE, a reinforcement learning framework that unifies these two forms of user representation by aligning language-based user profiles with embedding-based recommendation objectives. Given a user interaction history, BLUE leverages a profiler Large Language Model (LLM) to generate textual profiles, while an embedding model provides reward signals. This encourages the resulting textual representations to move closer to positive items and farther from negative ones in the embedding space. We further introduce a text-space supervision signal based on next-item prediction, ensuring the learned profiles remain both semantically meaningful and highly effective for downstream retrieval. Experiments on Amazon Reviews 2023 and Google Local Reviews in zero-shot sequential recommendation settings demonstrate that BLUE consistently outperforms strong baselines under both frozen and trainable embedding conditions. Notably, BLUE achieves clear gains in cross-domain transfer, highlighting the strong generalization ability of the learned user profiles. Furthermore, these generated profiles provide superior personalized context for question answering compared to raw user histories or alternative profile optimization methods. Overall, these results show that BLUE provides an effective way to unify interpretable textual profiling with discriminative latent embeddings for personalization.