Papers

Large language models are rapicly replacing search engines as the primary interface between people and information. Unlike search engines, which retrieve existing content, LLMs generate novel text shaped by internal representations learned during training. Here we show that partisan political identity is encoded in the model's activation space, and that this direction directly shapes generation. Using 190,491 tweets from sitting members of the U.S. Congress as labeled training data, we train linear probes on the hidden states of the Llama 3.1 8B Instruct model. We identify a single geometric axis at layer 18 that separates Republican from Democratic text with an AUC of 0.945 and a Cohen's d of 1.94, and use sparse autoencoders to decompose that axis into interpretable partisan features. Causally intervening along this axis, ablating or amplifying the partisan component mid-generation, produces systematic shifts in the model's output. We witness stance reversals, register shifting, and structured fabrications of authority. Our results demonstrate that partisan bias in language models is not a vague emergent property but a learned geometric feature that can be precisely located and steered. Partisan bias is not a bug to be patched, but a structural property of how these models encode information about their users. As LLMs displace search engines as the interface to knowledge, understanding that product design (and its consequences) will be essential for navigating the legal, social, and political transitions from an information ecosystem that is curated to one that is generated.

The analysis of internet memes in the Nepali language is complicated by frequent code-mixing and a lack of established baseline resources. While memes inherently combine visual and textual elements, this study focuses on a text-centric approach by extracting embedded text using an OCR layer and modeling it with Transformer-based architectures. We evaluate six distinct models and investigate the comparative effectiveness of Hard and Soft Voting ensemble strategies across two tasks: binary hate speech detection and three-class sentiment analysis. Experimental results show that a standalone decoder-only model achieved the highest performance for binary classification, whereas the Soft Voting ensemble performed best for the multi-class sentiment task, yielding a 15.8% relative improvement in Macro F1-score over the strongest standalone baseline. These findings suggest that ensemble strategies behave differently across binary and multi-class tasks, highlighting the importance of selecting aggregation methods suited to the classification objective.

Automatic evaluation is critical for high-stakes text generation, where errors often involve omitted findings, hallucinated content, polarity reversals, location changes, uncertainty mismatches, and temporal-comparison errors rather than low surface similarity alone. Radiology report generation provides a challenging test case because generated reports must preserve structured clinical evidence across sources. We present RadOT-Eval, an interpretable structured-evidence optimal transport framework for offline auditing of radiology report generation. RadOT-Eval decomposes reference and candidate reports into attribute-structured clinical evidence units, aligns corresponding evidence using entropy-regularized optimal transport, and uses clinically meaningful side-channel discrepancies in a monotone risk model to predict error burden. All transport, feature, and readout choices are selected using the ReXVal dataset, and the frozen system is evaluated on the independent RadEvalX dataset. RadOT-Eval achieves Spearman correlations of 0.715, 0.548, and 0.399 with total, clinically significant, and clinically insignificant annotated error burden, respectively, yielding higher point estimates than standard evaluation metrics and the open-source large language model (LLM)-based evaluator GREEN-radllama2-7B. In a frozen auxiliary corruption-sensitivity stress test on ReXErr-v1, RadOT-Eval achieves 0.768 AUROC and a 0.990 corrupted-greater-than-clean paired win rate. These results show that structured evidence transport provides an auditable, rank-oriented evaluation tool for high-stakes generated clinical text under ReXVal-only model selection and frozen RadEvalX testing.

Skill-augmented reinforcement learning improves language agents by storing reusable procedural knowledge acquired from past experience. Existing methods typically use strong language models to analyze trajectories, generate skills, and update a retrievable skill bank during online training. However, they rarely assess whether a newly generated skill is useful before it is stored and reused. We find that this assumption is unreliable: even skills generated by proprietary frontier LLMs exhibit highly mixed utility, with many providing little benefit or even degrading performance. Once such skills enter the bank, their effects are difficult to identify, because subsequent rollout feedback is delayed and usually reflects the combined effect of multiple retrieved skills rather than the marginal contribution of any individual skill. We propose an online reinforcement learning framework for pre-storage skill validation. The framework estimates whether a candidate skill contributes useful information beyond the skills already retrieved for the current task. It uses the standard rollout budget to form two matched groups under the same task and retrieval context: base rollouts conditioned on the currently retrieved skills, and skill-augmented rollouts conditioned on the same skills plus one candidate skill induced from the base trajectories. The reward gap between these two groups estimates the candidate skill's context-dependent marginal utility, enabling the framework to promote useful skills while filtering ineffective or harmful ones without additional rollout overhead. The framework further uses this marginal-utility signal to train the policy itself as a skill generator, reducing reliance on repeated calls to proprietary models. The learned skill-generation likelihood serves as a context-dependent score for retrieval-time reranking and outdated-skill pruning as the policy evolves.

We present HydraQE, our contribution to the IWSLT 2026 Speech Translation Metrics shared task. HydraQE is an end-to-end, reference-free quality estimation (QE) system for speech translation built on a Qwen3-ASR backbone, which accepts source audio and a translation hypothesis as joint input. Hidden states from all backbone layers are combined via a learnable sparsemax scalar mix, then re-encoded by a lightweight bidirectional Transformer to enable full cross-modal interaction prior to pooling into a shared embedding. Three independent prediction heads are trained on complementary supervision signals: human direct assessment (DA) annotations, MetricX-24 pseudo-labels, and xCOMET pseudo-labels. To address the scarcity of human-annotated data, we train on a combination of synthetically corrupted examples and silver pseudo-labeled machine translation outputs, using a curriculum that begins on synthetic and silver data and gradually shifts toward human-annotated examples. HydraQE outperforms cascaded text-based baselines and prior direct speech QE systems, demonstrating that end-to-end speech translation QE is competitive with cascaded approaches.

Mathematical reasoning has long served as a stringent test of machine intelligence; over the past decade, it has moved from a niche problem within NLP to one of the most consequential AI frontiers. This survey provides a unified account of the field's evolution, from early rule-based math word problem (MWP) solvers and template-driven geometry systems, through neural expression generation and LLM prompting, to contemporary reasoning models, multi-agent systems, neuro-symbolic theorem provers, and verified discovery workflows. We organize the landscape along four axes: (i) informal reasoning over text and diagrams, spanning MWP solving, multimodal geometry, and VLMs; (ii) formal reasoning in proof assistants, including autoformalization, tactic prediction, compiler-guided repair, and proof search; (iii) mathematical discovery, where systems propose constructions, improve bounds, or assist attacks on open problems; and (iv) the inference and training-time techniques, including CoT prompting, tool use, process reward models, and RLVR, that increasingly connect generation with verification. We catalog major benchmarks across grade-school arithmetic, competition mathematics, geometry, formal proving, multimodal and multilingual reasoning, and expert evaluation, and we examine benchmark saturation, contamination, reporting mismatches, and the distinction between pass@1, majority voting, and verifier-assisted pass@$k$. We critically assess failure modes: brittleness under perturbation, reward hacking, multimodal grounding failures, fragile formalization, and the energy cost of reasoning-scale inference. Drawing on recent perspectives from working mathematicians, we identify future directions centered on verified-discovery workflows, reasoning efficiency, and infrastructure to make AI-assisted formalization broadly usable. Companion materials: https://github.com/Starscream-11813/awesome-AI4Math.

Symbolic music evaluation for large language models remains fragmented across representations, datasets, and metrics. We introduce LilyBench, a LilyPond-based benchmark that jointly evaluates symbolic music generation and music understanding on the same family of open-weight LLMs. The benchmark includes a 200-prompt generation suite and ten understanding tasks adapted from ABC-Eval, covering syntax, metadata prediction, structural sequencing, and music recognition. Generation quality is evaluated using compile rate, MusPy descriptor distributions via Jensen-Shannon similarity, and LilyBERT-based Fréchet Music Distance (FMD). Experiments on four open-weight models show that executable LilyPond generation is achievable in zero-shot settings, while structural understanding tasks remain challenging despite strong performance on composer and genre recognition. Our experiments also reveal systematic disagreements between descriptor-based and embedding-based metrics, suggesting that symbolic music evaluation benefits from metric triangulation rather than single-score ranking. We release the benchmark, prompt bank, and evaluation code to support future research in symbolic music generation and understanding at https://github.com/CSCPadova/lilybench

We present a method for automatic Chinese web neologism detection that operationalizes traditional linguistic identification principles as prompt-engineering skills. The method has four stages: tokenizer-independent character n-gram candidate generation; dictionary anchoring with a Pointwise Mutual Information pre-filter; a well-formedness skill based on Chinese word-formation principles; and a combined rule and three-way classification skill that distinguishes neologism, entity, and none. Applied to the BAAI CCI 3.0 corpus (267M documents), the method produces 226,959 classified candidates including 4,853 labeled neologisms. To evaluate the method, we develop a per-stage conditional recall decomposition in which the pipeline's strict recall factors mathematically into the product of stage conditional recalls. Applied to Hou (2023) (4,199 entries), the decomposition exposes Stage 1 candidate coverage and Stage 4B LLM semantic judgment as the two bottlenecks (R=41.5% and 60.0% respectively), while intermediate stages are near-lossless. A length-stratified analysis further reveals that the structural well-formedness skill is length-invariant (>= 96.9%) whereas the semantic novelty-classification skill is length-dependent (65.6%/59.0%/44.1% across 2/3/4-character candidates), mapping a current boundary of skill-based linguistic operationalization. We release the method, pipeline outputs, and evaluation protocol as public resources.

Hallucinations – factually incorrect or unverifiable outputs – remain one of the most challenging limitations of Large Language Models (LLMs), especially in knowledge-intensive tasks. One proposed explanation is internal knowledge conflicts arising from fixed, outdated training data. This paper investigates whether internal representations linked to knowledge conflicts correlate with hallucination behaviors in LLMs. Using probing techniques inspired by two prior works, we analyzed activations from hidden, attention, and MLP layers, as well as output logits, across predefined tasks. We probed LLaMA-3-8B on hallucination detection benchmarks and Falcon-7B on a knowledge conflict dataset. Our findings show that, although conceptually related, hallucination activation patterns cannot be fully reduced to or explained by knowledge conflict representations. Nonetheless, probing proves a robust tool across multiple languages and activation types, supporting its role in improving LLM interpretability. This work advances the broader understanding of hallucinations in LLMs and underscores the value of fine-grained analysis of their internal behavior.

Pretrained models are often evaluated on multi-task leaderboards to measure their applicability in diverse contexts. However, current methods for aggregating performance across tasks into leaderboard-level rankings do not address the uncertainty and variability at the task level. While recent works have proposed interval-based model rankings, the principled aggregation of uncertainty from individual tasks to leaderboard-level rankings remains unaddressed, and variation in models' performance across tasks is frequently obscured. In this work, we introduce a hierarchical framework that constructs model rank intervals with statistical guarantees at both levels: task-level rank confidence intervals from pairwise comparisons, and leaderboard-level rank prediction intervals using a conformal approach. This enables reliable quantification of model rank for each observed task and for new potential tasks. Experiments on simulated data and the TabArena and PromptEval (MMLU) benchmarks show that our method yields statistically valid and informative intervals, enabling reliable, uncertainty-aware model ranking on leaderboards.

AI coding assistants have significantly improved developer productivity by automatically suggesting code that aligns with user intent, and many of these tools are now integrated directly into Integrated Development Environments (IDEs). Developers interact with code in two distinct cognitive modes: Flow and Command. While developers require tools that directly complete or infill code in unfinished programs during Flow mode, they also need tools that can comprehend intentions expressed as natural-language instructions and convert them into executable code in Command mode. Although instruction-tuned Large Language Models (LLMs) dominate many application scenarios due to their abilities to infer and fulfill developers' intents, it remains unclear whether the same paradigm is equally suitable for different code-related tasks. Therefore, it is necessary to understand how instruction tuning affects the feasibility of CodeLLMs as coding assistants. To fill this gap, we conduct the first empirical study that uncovers a key trade-off caused by instruction tuning across programming modes, which we term the Instruction-Tuning Tax. Our results show that instruction tuning is not a free lunch: although instruction-tuned models are more capable of following instructions and leveraging structured guidance, these gains often come at the cost of weaker infilling performance. We further extend our study through both qualitative and quantitative analyses, including manual failure categorization, behavioral metrics that capture generation fidelity, and intermediate-checkpoint evaluation throughout the tuning process. Summarizing our results into seven findings and four implications, our study offers a new perspective on the development of AI-powered coding tools and highlights the need to carefully balance instruction-following ability with effective code generation assistance.

Word-level cross-lingual alignment is central to annotation projection, translation auditing, and cross-lingual faithfulness estimation, yet existing neural aligners are rarely adapted to specialized domains. In this paper, we introduce ClinicalAligner26AM, a large-context multilingual aligner model for biomedical and clinical text initialized from ClinicalEncoder26AM. Our training recipe is inspired by AWESoME Align. We build our soft alignment target by sharpening with Sinkhorn-Knop optimal transport a cost matrix established for parallel clinical texts and conversations through the fusion of sentence-level, phrase-level, and token-level signals. We distill this sharpened alignment matrix directly into our student aligner, by encouraging its naive cosine-based token similarity scores to match this target. At inference time, we project source-span scores through the learned token alignment matrix and decode the longest valid high-scoring span in the target text, optionally supported by MultiClinNER predictions summarized in Appendix B. We evaluate CA26AM on the MultiClinCorpus shared task, which projects Spanish clinical entity annotations into six target languages. Our two submitted systems ranked respectively first and second across all languages and entity types, with character-weighted F1 scores above 0.95 in nearly all settings.

Large language model (LLM) agents are increasingly deployed in long-running settings where improving through experience at test time becomes important. A common approach is to update an explicit memory after each interaction to guide future decisions. However, most existing methods rely on hand-designed prompting rules, making it difficult to align memory updates with downstream objectives over multi-step horizons consistently. We propose MemoPilot, a plug-in memory copilot that explicitly trains the memory update process to improve a frozen LLM's performance across sequential interactions. We formulate memory updating as a multi-turn decision problem and optimize it end-to-end with multi-turn GRPO. Our training recipe introduces (i) a turn-wise reward signal and (ii) a context-independent, turn-level advantage estimation across rollouts, enabling finer-grained credit assignment and more stable training in multi-turn settings. We evaluate MemoPilot on two testbeds: multi-round Rock-Paper-Scissors (RPS) and Limit Texas Hold'em (LHE). Across both environments, MemoPilot substantially improves test-time learning of a frozen player over strong baselines, ranking first in Elo ratings on both games (1762 on LHE and 1590 on RPS) and outperforming all baseline memory methods and proprietary models, including DeepSeek-V3.2.

To interpret context correctly and retrieve relevant information, large language models must bind entities to their attributes and update these bindings as state changes. We analyze how LLMs implement this binding process in a dynamic state tracking. Using causal interventions, we identify a retrieval conditioned rebinding mechanism, a compact attention head circuit that propagated binding information and when the entity is queried, uses the updated binding to retrieve the corresponding attribute. Across Gemma and Llama models, this circuit supports rebinding behavior, but the representational signature of the mechanism differs across model families. In Gemma models, the binding signature is clearly expressed in the query/key subspaces of the relevant attention heads, whereas in Llama models, the binding information is carried primarily in key vectors. Overall, our results reveal an interpretable mechanism for context dependent state tracking in LLMs.

The increasing situational awareness of language models raises safety concerns: models might be aware when they are evaluated, and adjust their behavior to evade monitoring and resist modification, e.g., pretending to be aligned only in evaluation. This alignment faking behavior is often interpreted as scheming: an intentional effort of strategic deception. In this paper, we examine an alternative interpretation, performative misalignment, which explains the change in behavior as a result of sycophancy towards AI researchers. To examine this hypothesis, we present three empirical findings. First, we show that evaluation awareness persists even when we tell models they are deployed, which contradicts the scheming story which predicts less misalignment when the model perceives evaluation. Second, we use probing and steering to show that our current methods cannot mechanistically distinguish sycophancy and scheming in alignment faking evaluations. Third, we fine-tune models to be more sycophantic and observe increased sensitivity to evaluation cues. To conclude, we emphasize deconfounding sycophancy from scheming for future work on evaluations and mitigations of intent misalignment.

As Large Language Models (LLMs) advance toward open-ended autonomous agents, the mechanisms used to evaluate and guide their behavior must evolve accordingly. This work introduces the rubric as a unifying framework capturing this evolution, characterizing rubrics as a dynamic response to successive LLM paradigm shifts that recurs across otherwise independent efforts in evaluation, reinforcement learning, and safety alignment. We define rubrics as explicit criteria sets that transform complex quality judgments into structured and actionable standards, and demonstrate that their recurrence across these research threads is not coincidental. We systematically organize existing rubric designs, examine their construction and optimization, and analyze their role across evaluation and training. Rubrics manifest at three progressively deeper levels: at the evaluative level, they decompose holistic judgments into verifiable dimensions; at the training level, they serve as dense feedback signals providing process-level guidance where scalar rewards fall short; at the intrinsic level, they emerge dynamically from model behaviors, driving self-improvement. We further assess rubric reliability across generation quality, execution fidelity, theoretical constraints, and security threats, before surveying rubric-based benchmarks across diverse domains. By rendering assessment transparent and decomposable, rubrics translate human value expectations into machine-learnable signals, serving as the enduring bridge between human intentions and machine behavior.

Author name disambiguation is a critical challenge in academic search systems, often addressed through from-scratch and real-time disambiguation approaches. However, current algorithms remain vulnerable to cumulative errors of paper-author assignments and overlook inconsistent assignments across different sources. Resorting to expert annotation is resource-intensive. To this end, this paper explores a new perspective for author name disambiguation: cross-source correction by leveraging inconsistent assignments across sources. We propose CrossND, a full-stack framework that integrates data refinement, cross-source reasoning, and test-time scaling. First, a chain-of-refinement pipeline denoises author profiles and produces more accurate paper-author matching probabilities. Second, a supervised fine-tuning process incorporates these refined signals and a probabilistic soft logic-based cross-correction module to infer the assignments of which sources are incorrect. Third, test-time scaling further enhances the accuracy and robustness of the predictions. Experiments on real-world datasets indicate that CrossND consistently outperforms 17 baselines by leveraging cross-source reasoning without human intervention.

Vision-Language Models (VLMs) are increasingly required to process unbounded video streams in applications such as video-call assistants, live commentary, and embodied robots. An ideal streaming system should support proactive interaction, long-horizon memory, and real-time processing, while resting on a VLM backbone capable of handling diverse in-the-wild streaming tasks. However, existing VLMs excel at offline video understanding but fall short in streaming capabilities and lack dedicated infrastructure for streaming deployment. We address this gap on three fronts. (i) For backbone capability, we construct Streaming-Train-248K, a streaming dataset paired with a novel training objective for adapting VLMs to streaming interaction and understanding. (ii) For real-world deployment, we introduce Streaming Harness, a plug-and-play system that endows any VLM with three core abilities: proactive interaction (per-second response decisions), long-term memory (12-hour context retention), and real-time processing (sub-second latency). (iii) To drive continued community progress on streaming capabilities, we design Streaming-Eval, a benchmark that reflects models' capabilities across diverse in-the-wild scenarios. Extensive experiments demonstrate consistent gains from our approach across all core capabilities required for streaming video understanding. We will open-source our data, code, and benchmark to advance the community's shift from offline video understanding to deployable streaming intelligence.

We present a multilingual fact-checking system deployed at Factiverse, designed for high-throughput and low-latency operation across diverse languages. The system follows a modular pipeline with three stages: claim detection, evidence retrieval and re-ranking, and veracity prediction. We fine-tune XLM-RoBERTa-Large for claim detection, mmBERT-base for three-label stance classification (Supports/Refutes/Mixed), and a SetFit-based multilingual re-ranker for claim–evidence matching. We compare these components against strong LLM baselines, including GPT-5.2, Claude Opus~4.6, and Qwen3-8b. Experiments on production data spanning 114 languages for claim detection and 28 languages for veracity prediction show that task-specific fine-tuning provides strong and stable multilingual performance, while the fine-tuned retrieval model remains competitive with modern proprietary embeddings. Same-hardware latency measurements further show large efficiency gains for encoder-based components, supporting their use in production deployments with tight cost and privacy constraints. Overall, compact fine-tuned, self-hosted models remain a practical and effective foundation for multilingual fact-checking at scale. Code and data used for this study are available at https://github.com/factiverse/factcheck-editor.

Purpose - Quotation error refers to the inconsistency between cited information and its original source. This phenomenon leads to a series of negative impacts, such as misinterpretation of the original research, undermining the academic community's collective understanding of relevant issues, and weakening the accuracy and fairness of the citation-based academic evaluation system. Existing studies have shown that quotation error is prevalent in the academic community; moreover, manual verification of quotation error is not only labor-intensive but also inefficient. Therefore, this paper proposes the task of 'automated detection of quotation errors'. Methodology - Adopting a large language model (LLM)-based approach, this paper improves detection performance from two aspects on the basis of existing research: first, employ the fine-tuning approach for LLMs to detect quotation errors; second, incorporating full-text data of the cited literature into dataset construction, and exploring the optimal scheme for building such datasets by comparing three types of full-text integration methods. Based on this, this paper further uses the TokenSHAP tool to conduct interpretability experimental analysis on the model's prediction results. Findings - The fine-tuning approach for LLMs has improved the performance in detecting quotation errors. Among the different methods for incorporating full-text information, the approach based on using the source abstract yielded the best performance. Originality - The fine-tuning approach for large language models (LLMs) is applied to the task of automated detection of quotation errors, and interpretability analysis is conducted on the model's output results.

Speech emotion recognition (SER) is commonly formulated as utterance-level classification, although conversational emotion depends on a speaker's usual vocal range and the emotional context established by previous utterances. Speech-language models provide strong pretrained acoustic and semantic representations, and can adapts them to SER labels via finetune, but this mechanism still missing per-dialogue state. We study whether test-time neural memory can supply this missing context while leaving the large audio language models (LALMs) backbone intact. Building on Titans, we introduce a plug-and-play Memory-as-a-Layer (MAL) adapter that writes dialogue history into a small neural memory and reads it back as an audio-token-aligned residual update, avoiding changes to the host model's token positions. Across different audio LLMs and emotion recognition datasets evaluations, our design improves SER performs across different evaluation metrics, supporting test-time memory as a residual contextual mechanism for conversational SER.

Large language models frequently fail in a characteristic way: rather than acknowledging ignorance, they produce fluent but incorrect answers to questions that lie beyond their knowledge boundaries. We introduce Structured Ignorance Certificates (SICs), a JSON-formatted output schema that demands a model explicitly name the missing domain intersection, enumerate required concepts, and propose a productive retrieval query rather than hallucinating an answer. To train models to produce high-quality SICs we construct a 7,347-sample Unknown-Unknown (UU) dataset by prompting Qwen3-14B to stitch together questions from seven domains (physics, biology, engineering, CS, economics, medical, legal) into novel cross-domain queries that no single-domain expert could answer. We fine-tune a 14B-parameter model with Group Relative Policy Optimization (GRPO) using a composite reward that combines retrieval utility, concept specificity, and output-format validity. A paraphrase-divergence probe trained on model responses confirms that SIC-tuned outputs systematically exhibit higher unknown-unknown probability scores. Evaluation on 735 held-out UU questions achieves a 99.46% JSON validity rate, a mean Certificate Specificity Score of 0.967, and a 3.6% ROUGE-L improvement over the base model on retrieval-grounded generation – demonstrating that explicit epistemic structuring is a learnable and measurable capability.

Transformer language models process input provided as subword fragments, but natural language semantics usually rely on word-level concepts. Detokenization is the process where models reconcile these two facts, aggregating subwords into word-level representations through their computation. Prior work has found that this takes place mostly in early-to-middle layers, but so far the exact mechanics of the process have not been pinned down. We venture deep into detokenization using activation patching in controlled paired experiments that isolate the contribution of different model components, localizing English detokenization in Llama2-7B to a two-stage process at Layer 1. Attention transmits a token-specific signal from nonfinal subwords, using sequential relays if necessary, while the MLP composes it with the local embedding. This two-stage structure generalizes to twelve models from eight families, but the depth over which it takes place depends on the flavor of positional encoding: RoPE-based models detokenize over 1 to 5 layers, while learned-absolute models take 5 to 10. Finally, we provide a probe for determining the success of the detokenization process based on early-layer activations alone, performing at 0.94-0.97 AUROC depending on the amount of context.

Building Information Modeling (BIM) projects require information requirements to be described as machine-checkable Information Delivery Specification (IDS) files in order to verify whether building models contain the required attributes. However, IDS authoring remains a practical bottleneck: practitioners must handle domain vocabulary, strict XML schema constraints, and external validator conformance while also checking whether the requirement itself is correctly expressed. We present Ishigaki-IDS, an open-weight LLM specialized for verifier-aware IDS draft generation. The model combines continued pretraining on BIM/IDS corpora, supervised fine-tuning on information-requirement-to-IDS pairs, and reinforcement learning with verifiable rewards from an external validator. The goal is not to replace expert review, but to move IDS authoring from low-level XML and schema repair toward validator-loadable drafts that practitioners can inspect and correct. On the 166-case expert-created Ishigaki-IDS-Bench, Ishigaki-IDS-8B achieves an IDSAuditPass score of 0.651, a validator-pass metric for generated IDS files, substantially outperforming Claude Opus 4.5, the strongest single-shot LLM baseline we evaluated, at 0.331. It also obtains an Audit-Gated FacetF1 of 0.282, which measures requirement-facet alignment among validator-passing drafts. The same recipe scales: 14B and 32B variants reach IDSAuditPass 0.753 / 0.693 and Audit-Gated FacetF1 0.392 / 0.369. In a workflow check with six BIM practitioners, Ishigaki-assisted authoring reduced aggregate work time by 54.7% under the same validation and alignment endpoint. These results suggest that verifier-aware IDS generation can reduce the practical burden of converting BIM information requirements into reviewable IDS drafts.

Published agent capability scores conflate what a model can do with what its scaffold lets it do, and the magnitude of this elicitation gap is not well characterized under controlled conditions. This study executes a pre-registered controlled comparison of three scaffolds (ReAct, a Planner-Actor-Rater multi-agent design, and planner-then-executor) across five models from three providers (Claude Opus 4.7, Sonnet 4.6, Haiku 4.5; Gemini 3.1 Pro Preview; GPT-5.5) on GAIA validation Levels 1 and 2, holding tasks and conditions fixed, with three attempts per question. Scaffold choice alone moves measured accuracy by as much as 28 percentage points within a single model (Opus, Level 2, robust slice), confirming the pre-registered hypothesis that scaffold variation produces gaps of at least 10 points. The pre-registered prediction that more capable models would be less scaffold-sensitive is rejected in direction: scaffold effects vary significantly by model in every dataset slice, but the most capable Anthropic model gains the most from structured scaffolds at the harder level, and tier-scaling holds only at Level 1 under the robust slice. The multi-agent advantage over ReAct at Level 2 appears within the Anthropic family but not for the cross-provider models, making model family rather than capability tier the conditioning variable, and the predicted planner-executor advantage on file-reading tasks is falsified. Structured scaffolds make fewer tool calls yet recover more often from mid-trajectory errors at the harder level, and a single cell (Gemini with planner-then-executor) is the cheapest at both levels and the most accurate at Level 2. These results indicate that single-scaffold capability numbers are scaffold-conditional estimates and that the elicitation gap is not guaranteed to shrink as models improve.

Selective predictors answer on confident inputs and abstain elsewhere; deploying one safely needs a single finite-sample certificate that simultaneously upper-bounds the selected risk, lower-bounds the acceptance probability $\pacc$ above a floor $\pmin$, and lower-bounds the deployment utility. This certificate must be valid under adaptive threshold selection from a finite grid of $m$ pairs on $\ncert$ samples. We give such a certificate for bounded, possibly non-monotone losses by treating the selected risk directly as a ratio rather than through a Hoeffding-style range bound. The construction couples three confidence bounds: a variance-adaptive empirical-Bernstein bound on the ratio risk, a Clopper–Pearson bound on acceptance, and a two-sided closeness bound on utility. Together they lower-bound the certified policy's utility absolutely and to within $2\gammau$ of the best over the certified set, both non-vacuous whenever feasible; a regime-scoped third leg matches an external oracle, informative only where the risk margin $\gammar < \alpha$ and vacuous at the headline operating points. Relative to the range-only Hoeffding-ratio construction this sharpens the acceptance-floor dependence from $1/\pmin$ to $1/\sqrt{\pmin}$, and a closed-form corollary identifies a per-pair regime in which our risk bound dominates a Hoeffding conformal risk control (Hoeffding–CRC) selective bound. Empirically, on ImageNet (three ResNets) and COCO val 2017 panoptic, the certificate opens a $+22$ pp certified-acceptance frontier over Hoeffding–CRC and is ${\approx}10{\times}$ tighter than a non-vacuous matched-valid baseline; these gains are regime-scoped, not universal, and absent on ADE20K. The certifier runs in $O(\ncert m)$ time.

As Russia's war against Ukraine extends into generative AI, large language models (LLMs) adapted for local post-Soviet languages are deployed in contested information environments. Policy and industry discourse assumes that culturally aligned adaptation encodes the political orientation of the target community: a Ukrainian-oriented model will resist Russian narratives, a Russian-oriented one will reinforce them. Does it? This article systematically disconfirms that assumption. We run a controlled audit of four openly available LLMs sharing a common base model but fine-tuned for different linguistic communities, querying them in Ukrainian, Russian and English across ten contested wartime narratives: Crimea, "denazification", the "one people" thesis, and atrocity denial at Bucha and Mariupol. The result is a Fine-Tuning Paradox: the Ukrainian-oriented model shows the weakest resistance to Russian disinformation in Russian, while the Russian-oriented one exhibits the strongest rejection. Corpus composition, language coverage and prompt format prove more decisive than nominal cultural provenance. We situate these findings within debates on hybrid warfare, digital sovereignty and post-imperial information orders, arguing that the principal threat to regional information sovereignty is not adversarial fine-tuning but the untested assumption that cultural alignment guarantees resilience.

Reinforcement learning (RL) holds immense promise for enhancing the reasoning capabilities of diffusion large language models (dLLMs). However, progress is fundamentally constrained by a dual misalignment between authentic generation trajectory and the gradient update process: (i) Process-reward misalignment. Sparse, terminal rewards are indiscriminately assigned to all intermediate steps of the generation process, failing to provide discriminative credit assignment. (ii) State-trajectory misalignment. Policy updates are often diverted toward artificial, out-of-trajectory states, squandering gradients on less informative samples. To address these limitations, we introduce Process Aligned Policy Optimization (PAPO), a novel framework that holistically aligns the RL update with the dLLM's generative trajectory via Step-Aware Process Rewards (SPR) that transform sparse terminal rewards into dense, step-wise credit, and Entropy-Guided Historical Re-enactment (EHR) that replays authentic trajectories at high-uncertainty steps. Extensive experiments on four benchmarks demonstrate that PAPO significantly outperforms baselines, achieving gains of 4.5% on GSM8K, 4.8% on MATH500, 42.2% on Countdown and 16.1% on Sudoku.

As deep language models (DLMs) are increasingly deployed in high-stakes domains such as healthcare, understanding their decision rationale becomes paramount for ensuring trust, safety, and accountability. However, achieving this vital level of interpretability is particularly challenging when these DLMs operate as black-box systems (e.g., via APIs), where access to internal model states (e.g., parameters, gradients) is restricted. Despite numerous efforts, existing explanation methods often fail to concurrently satisfy three key desiderata: (i) inference-time efficiency, (ii) black-box compatibility without inducing out-of-distribution behavior, and (iii) comprehensible explanations grounded in the input's linguistic structure. To address these challenges, we propose a method that explains predictions of DLMs by selecting a small, informative subset of input words. We formulate this as an amortized optimization problem, enabling efficient one-shot inference without the need for input-specific search. Our selection policy is trained via REINFORCE-style policy gradients, allowing discrete word selection in a fully gradient-free setting. To enhance interpretability and align with human linguistic intuition, we integrate graph-structured knowledge into this selection process, fostering linguistically coherent subsets that result in explanations both highly informative and cognitively meaningful to end-users. We evaluated our method on diverse DLM architectures and multiple real-world datasets. It consistently identifies word subsets with enhanced discriminative power and stronger alignment with linguistically salient cues, outperforming both conventional black-box compatible methods and gradient-based approaches that are given oracle access to the black-box model's gradients for a more challenging benchmark. Our code is available at here.

Although Sparse Autoencoders (SAEs) have mitigated the opacity of large language models (LLMs) by decomposing dense representations into sparse features, explaining these features still remains a central challenge. Current explanation methods, however, typically operate within an open-loop paradigm, failing to leverage mechanistic feedback for further refinement. In this paper, we propose SAEExplainer, a training framework that utilizes activation scores as an objective reward signal to train the model for self-correction and iterative bootstrapping. By iteratively verifying and correcting foundational explanations through a two-round optimization process, SAEExplainer achieves continuous improvement in its explanatory capabilities. This mechanism significantly reduces explanation hallucinations and reinforces causal triggering patterns. Extensive experiments demonstrate our approach improves upon established baselines across most metrics, especially in causal triggering and discriminative activation. The code is available at https://github.com/he-jingyi/SAEExplainer.