Papers

Coding agents have made substantial progress on software engineering tasks that modify existing codebases, including bug fixing and feature implementation. However, constructing a complete program from scratch remains a major challenge: even the frontier models evaluated on ProgramBench fully resolve fewer than 1% of tasks. One obstacle is the lack of scalable training environments for this from-scratch setting, spanning the whole software engineering life cycle, as existing environment-construction frameworks focus only on a single phase in software development. To address this gap, we introduce MindForge, an automated pipeline that converts open-source command-line programs into source-free environments that expose only a compiled reference executable and its documentation. Using MindForge, we construct training environments from repositories disjoint from those in ProgramBench, and curate a high-quality data recipe consisting of program synthesis trajectories using GLM-5.2 as the teacher agent. Fine-tuning Qwen3.6-27B on these trajectories increases its ProgramBench average test pass rate from 37.98% to 49.51%, achieving performance comparable to substantially larger frontier models. Moreover, the fine-tuned model consistently improves over the base model across all seven unseen software engineering benchmarks, spanning long-horizon repository generation and translation, bug fixing, feature implementation, and cross-language issue resolution, with absolute gains of 31.00 points on RepoZero-C2Rust, 14.16 on DeepSWE, 10.70/4.56 on NL2Repo-Bench (with/without tests), 5.04 on SWE-bench Verified, 5.93 on SWE-bench Pro, 5.22 on SWE-bench Multilingual, and 4.94 on FeatBench.

Writing and communication are increasingly mediated by large language models (LLMs) that are being used to draft, revise and polish text. Although such assistance can improve clarity and help authors meet institutional expectations, widespread reliance on shared models may reduce population-level variation in linguistic form, a phenomenon we refer to as linguistic monoculture. We develop a mathematical framework in which authors and LLMs are represented as distributions over linguistic features and coevolve through repeated interaction. We analyze three interaction mechanisms: a shared model with a fixed linguistic distribution, a shared model recursively updated from author outputs, and personalized models updated through author-specific and population-level feedback. We characterize the resulting equilibria and convergence rates, showing that, shared models can drive authors toward a common norm, recursive feedback relocates the shared norm without altering pairwise spread under common conformity, and personalization can preserve a family of distinct author-model equilibria with nonzero linguistic diversity. We then endogenize conformity as a strategic choice trading off private benefits from clarity, legibility, and perceived fluency against distinctive style. Within this utility model, individually rational authors may conform more than is socially optimal because they do not internalize the value their distinctiveness provides to others, creating a negative externality and a price of monoculture that is finite for each fixed instance but can grow without bound when distinctiveness dominates authenticity. Synthetic simulations illustrate how fixed shared assistance, recursive feedback, and personalization produce different long-run diversity outcomes.

Fine-tuning is the dominant paradigm for specializing large language models (LLMs), yet it exposes a critical vulnerability: malicious data providers can embed harmful behaviors into downstream corpora, creating models that retain professional skills while violating human values on demand. Existing safety-realignment defenses often fail in practice due to three key limitations: they frequently cause catastrophic forgetting of specialized skills; their effectiveness collapses when the defender cannot observe the attacker's prompt template; and successfully realigned models remain susceptible to re-jailbreaking via simple system prompt switches. To address these challenges, we propose Routing-based On-Policy Distillation (ROPD), a novel realignment framework that models the divergence between aligned and compromised output probability distributions rather than fitting specific prompt templates. We conduct extensive experiments comparing ROPD against four state-of-the-art baselines across three datasets and three base models with varying alignment strengths. Our results demonstrate that when baseline defenses face template mismatches, often accompanied by severe degradation in downstream task performance. In contrast, ROPD substantially mitigates template-mismatch risks, maintaining superior robustness in both defense effectiveness and capability preservation. While our analysis indicates ROPD is not entirely immune to template shifts, its performance degradation is negligible compared to existing methods, establishing a new standard for robust LLM realignment.

Personalized agents are increasingly applied to assist users across a wide range of tasks. Effective personalized assistance requires not only retrieving explicit facts from past interactions stored in agent memory, but also inferring abstract personal characteristics. However, existing memory benchmarks primarily evaluate whether an agent can retrieve information explicitly stated in conversational histories, failing to provide an effective assessment of deeper user understanding. In this work, we propose Setoka, a benchmark for evaluating memory-augmented personalized agents with hierarchical user understanding from heterogeneous data. Grounded in theories from cognitive and personality psychology, Setoka defines four levels of user understanding, i.e., semantic memory, episodic memory, behavior pattern, and personality trait. Moreover, to enable realistic yet privacy-preserving evaluation, we design a psychometrics-based pipeline that synthesizes diverse, coherent heterogeneous user data and queries at scale. Finally, we leverage Setoka to evaluate 3 language models combined with 5 memory systems for 10 synthetic users. Our comprehensive evaluation reveals that while existing systems perform well on semantic memory retrieval, their performance declines on episodic memory. Moreover, when dealing with behavior pattern and personality trait understanding tasks that require integrating heterogeneous and fragmented information dispersed over time, performance declines even further. These findings demonstrate that user understanding cannot be handled by simple fact retrieval, motivating the design of memory mechanisms for cross-source integration and abstraction over long-term user behavior.

Regional bias in large language models (LLMs) may shape both perceptions of regional groups and decisions about individuals from different regions. Yet existing studies often examine these manifestations separately, leaving their structure and consequences unclear. We introduce Stereotypes-to-Decisions (S2D), a systematic framework evaluating regional bias from abstract stereotypes to concrete social decisions. Covering all 34 provincial-level administrative regions of China, S2D evaluates six LLMs using stereotype ratings of Warmth (perceived friendliness and trustworthiness) and Competence (perceived capability and intelligence), along with paired-choice tasks across Education, Occupation, and Social Interaction. Results reveal substantial regional differences in regional scores, with considerable agreement across models, especially for Competence and Occupation decisions. Furthermore, these patterns are associated with regional economic and digital development indicators and display mixed human-like stereotypes, with some regions rated highly on one dimension but poorly on the other. They also remain largely stable across Chinese and English prompts. Overall, our findings show that regional bias in LLMs is prevalent, systematic, and consequential, motivating more regionally aware evaluation and mitigation.

Large language model (LLM) agents automate penetration testing through an observation-action loop, selecting actions based on observations returned by tools. This dependence allows defenders to inject deceptive observations that can mislead the agent's decision-making process. However, existing defenses rely heavily on static, isolated artifacts planted in the environment prior to an attack. Advanced agents can progressively recognize and bypass these artifacts, ultimately refocusing their exploitation attempts on the real target. To address this issue, we introduce AgentSnare, a trajectory-adaptive deception system that dynamically unfolds a decoy environment to continually steer the penetration agent away from the real target. Specifically, AgentSnare employs an artifact-construction policy model that constructs candidate artifacts conditioned on the agent's interaction history and decoy state. AgentSnare then validates these candidates and incrementally incorporates valid artifacts into a factually consistent decoy environment, thereby delaying the attack by absorbing its tool calls, diverting its post-entry trajectory within the decoy, and defusing it by inducing completion reports grounded in decoy evidence. Across 15 CVE-Bench web applications and three attacker models, AgentSnare absorbs 46.8% of the agent's tool calls in the decoy and retains 55.9% of post-entry actions there, while 90.0% of completion attempts are grounded in decoy evidence; across all 45 attacker-CVE pairs, no real target is successfully exploited at pass@3.

Large language models are increasingly used as decision aids whose probability judgments shape downstream choices. Whether those judgments carry a systematic directional tilt has been hard to detect: calibration metrics aggregate unsigned errors, and naturalistic uncertainty offers no ground-truth probability. When an LLM rates a startup's success at 70% but its failure at 15%, the missing 15 points expose a distortion no aggregate score flags. We introduce OptimismBench, which detects directional bias with inverted pairs: each scenario elicits both P(success) and P(failure), and asymmetry between the two framings yields a signed bias score without ground truth. Across 16 models from 8 providers, fourteen are optimistic; pessimism appears only in Anthropic's frontier tier. Eleven matched base-versus-chat pairs across four families show post-training sets the sign of the bias, with opposite shifts in different families. The pattern survives prompt, temperature, perspective, and self-debiasing ablations. A seventeen-model six-language comparison further shows model identity dominates language, with inter-model variance at 4.7x inter-language variance. We release 3,870 items across 10 languages for per-model directional-bias auditing. When alignment makes a model more helpful, it also tilts its probabilities; downstream pipelines inherit the tilt by default.

Travel planning is a demanding stress test for tool-using LLM agents: a usable itinerary is a single artifact that must be right along many axes at once - every flight, hotel, and attraction must exist and be bookable, the days must be physically traversable, the total must clear a budget, and the plan must serve a traveler whose needs are only partly stated. Existing agent benchmarks reward these properties one at a time and grade the final output with soft or LLM-judged rubrics, which cannot certify that a returned plan is executable and are neither reproducible nor auditable. We introduce TREK (Travel Reasoning and Evaluation Kit), a benchmark for feasible itinerary synthesis: producing a single plan that is jointly constraint-correct, hallucination-free, spatio-temporally executable, budget-valid, and responsive to the traveler's unstated persona needs. TREK comprises 800 multi-constraint tasks - 533 feasible and 267 provably infeasible with typed route/entity/budget causes - over a synthetic, internally consistent knowledge base of 212,530 records across 375 cities and 13 personas, served through a production-style tool sandbox of validated RESTful APIs. Every task is scored by a fully deterministic, rule-based evaluator with no LLM judge and ships a human-verified gold reference that scores a perfect 1.0 under that same evaluator, so the ceiling is demonstrably achievable and every remaining gap is an agent limitation rather than scorer strictness. Evaluating 15 LLM agents across nine constraint dimensions, we find that even the strongest (GPT-5.6) produces a fully-feasible plan on only 46.2% of solvable tasks, with a median of 6.6% and a floor of 0.0%; satisfying travelers' unstated needs emerges as the universal bottleneck, unsolved even at the frontier. We release the dataset, tool sandbox, deterministic evaluator, and agent code as a fully reproducible benchmark.

Emotion-cause pair extraction in conversation (ECPEC) identifies utterance pairs in which one utterance causes an emotion expressed in another. Recent LLM-based approaches formulate ECPEC at markedly different granularities, ranging from generating complete pair sets to judging individual candidate pairs. In this paper, we make the surprising observation that task formulation substantially affects performance, where pair-level judgement outperforms dialogue-level generation in all 18 controlled comparisons. We investigate the sources of this paradigm gap and find that many relations omitted by dialogue-level generation remain recognizable under explicit pair queries, under which the model recognizes 92.7%-98.1% of emotion-cause relations. This suggests that LLMs can recognize emotion-cause relations but struggle to discover and return complete pair sets. Pair-level judgement alleviates this burden, although its candidate rankings are more reliable than the binary decisions produced by a shared threshold. Based on this diagnosis, we introduce an auxiliary retriever that selectively re-examines ambiguous boundary cases, yielding consistent F1 improvements of 0.50-1.46 points across three datasets while maintaining an inference time of only 1.49x that of the baseline paradigm. These findings show that task decomposition and candidate scope are critical to effectively utilizing LLMs for ECPEC.

We introduce CreditCardQA, the first financial literacy benchmark for numerical reasoning derived from real credit card agreements. The dataset contains 1,800 questions, including first-person variants that reflect how consumers naturally ask about fees, interest, and payments. We evaluate a range of large language and reasoning models under Chain-of-Thought (CoT) and Program-of-Thought (PoT) prompting. Overall, PoT yields consistent performance gains, particularly for models with weaker baseline reasoning, and narrows gaps between open- and closed-source systems. Through error analysis, we show that failures arise less from arithmetic and more from misapplied financial rules, missed conditions, and misunderstandings of contractual terms. We further analyze question difficulty and find that comparisons, conditional logic, and monetary constraints are especially challenging. We also find that errors often arise in edge cases such as late-payment penalties or small-balance scenarios that are more likely to affect lower-income or financially vulnerable individuals.

The same diagnostic result can support or challenge one causal claim yet fail to address another when the claims concern different populations, outcomes, estimands, pathways, or identifying assumptions. When the evidence and target vary together, a correct answer may reflect favorable or adverse wording, lexical overlap, or a familiar diagnostic pattern rather than matching the evidence to the causal question. We introduce paired prompts that repeat the same diagnostic evidence verbatim while changing the causal target. Each prompt is labeled Favors, Challenges, Unresolved, or Wrong Target according to how the evidence bears on the causal question. A pair is recovered only when both prompts are classified correctly. Using linear readouts trained on a separate development set, we analyze the final-token hidden state from the penultimate transformer block of Qwen2.5-7B-Instruct, Qwen3-8B, and Llama-3.1-8B-Instruct. On the 49-pair primary benchmark spanning nine diagnostic families, balanced accuracy ranges from 0.654 to 0.659 and 18-21 pairs are recovered. Two independent human reviewers assigned the same label to 95 of the 98 prompts (96.9%). Across checkpoints, balanced accuracy and complete-pair recovery exceed permutation nulls that preserve development scenario groups. In Qwen2.5, full-prompt balanced accuracy exceeds both restricted inputs, with paired-bootstrap intervals for both differences above zero. Readouts trained without development examples from the evaluated diagnostic family recover 21 pairs, including at least one in each of the nine families. The hidden-state readout exceeds a linear classifier on answer-option logits and text baselines in balanced accuracy and recovered pairs. These results show that the hidden state contains linearly decodable information about whether diagnostic evidence favors, challenges, or fails to address the causal target.

Classical spoken dialogue systems often separated dialogue management from response realization: a policy selected the next dialogue action, and a generation component expressed that action. As dialogue systems shift toward LLMs, this decomposition has largely disappeared into the model's hidden representations. We ask whether an LLM-internal analogue of state estimation and action control can be recovered for conversational moves such as acknowledging, checking, querying, explaining, and replying. We formulate move control as two coupled problems: selection, predicting the appropriate next move from the dialogue context, and realization, causally producing a chosen move at generation time. We introduce Latent-IM, an internal dialogue-management framework that provides a general interface for choosing and deploying conversational moves under different objectives. Here, we use this control to reproduce human move choices, improving average end-to-end move accuracy by 12.5 points over the unsteered backbone while performing comparably to fine-tuning.

Knowledge graph completion (KGC) aims to infer missing facts in knowledge graphs (KGs), thereby improving their completeness and supporting downstream intelligent applications. However, emerging entities and relations in real-world deployments make inductive KGC difficult, especially under few-shot and zero-shot settings. Multimodal information and Large Language Model (LLM)-derived priors can enrich sparse relational contexts, but they may also introduce noisy or hallucinated evidence. To address these issues, we propose DuPLeR, a Dual-Path LLM Reasoning framework for multimodal few-shot KGC. DuPLeR builds a calibrated relation graph by combining multimodal LLM-derived type priors with factual support structures, and performs dual-level structural reasoning over the refined relation topology. Moreover, a dual-pathway multimodal enhancement module regulates message passing with query-relevant multimodal signals and supplements entity representations after graph propagation. Experiments on eight inductive variants of two multimodal KG (MMKG) benchmarks show that DuPLeR achieves robust performance in data-scarce KGC scenarios.

Sequence labeling is a fine-grained information extraction task, yet existing large language model-based approaches suffer from insufficient domain alignment and low inference efficiency. To address these issues, we propose DIRECT, a framework that addresses these issues through training-time optimization and inference-time rectification. Specifically, DIRECT performs Direct Preference Optimization (DPO) after supervised fine-tuning to strengthen task alignment with human preferences, and introduces a controlled decoding process that enforces fixed output formats and restricts predictions to candidate sets. To further improve efficiency, a template-filling mechanism requires the model to generate only label tokens while reusing prefixed content through the KV Cache, thus reducing redundant computation. Experimental results on eight datasets demonstrate that DIRECT achieves significant improvements in both performance and efficiency compared to existing methods.

When asked to describe a medical image that was never attached, frontier vision-language models do not abstain: they confabulate a diagnosis. We show that this confabulation is not random. It is structured by who the patient is said to be. Across chest X-ray, brain MRI, and dermatology, Claude Opus-4.7, GPT-5.4, and Gemini-3.1-Pro are each queried with only a demographic descriptor and no image, and changing the descriptor systematically shifts the diagnosis returned. Claude concentrates sharply: a 65-year-old white man asking about a skin mole receives Melanoma in nearly every response, and a 32-year-old Black woman asking about her chest X-ray receives a Sarcoidosis diagnosis whose reasoning reads "suspected, based on demographics and classic pattern.” GPT-5.4's effect is broader, fabricating across every demographic cell we test, most conspicuously naming Sarcoidosis for young Black patients on chest X-ray. Two structural findings sharpen the problem. A hedged regime appears in which the prose acknowledges the missing image while the structured diagnosis field nevertheless names a disease, a dissociation invisible to prose-only audits. And Claude's dermatology effect collapses entirely when 'skin mole' is swapped for 'skin lesion' while GPT-5.4's is preserved, indicating that mirage is a family of distinct failure modes rather than a single phenomenon. Trustworthy VLM deployment in clinical pipelines requires auditing the structured output channel directly, and probe-word sensitivity should be treated as a first-class evaluation dimension

Test-time reinforcement learning (TTRL) enables language models to self-evolve at inference time without labeled feedback. Existing methods rely on answer voting and therefore do not extend naturally to open-ended generation, where valid responses cannot be mapped to a shared canonical answer. Without external reward models or stronger judges, adaptation must instead construct reliable rewards from the model's own outputs. We introduce SERPO (Self-Evolving Rubric Policy Optimization), which replaces answer voting with a closed loop that co-evolves response evidence, query-specific rubrics, and policy parameters. Good-Normal-Bad (G-N-B) response evolution organizes maximally separated rollouts into ordered archives; rubric evolution retains criteria that discriminate these archives; probabilistic criterion scoring converts verdict-token likelihoods into reward signals; and policy evolution optimizes the actor with the resulting signals. New actor rollouts then refresh both the archives and rubrics, closing the three-way evolution loop. Across two model configurations, two in-domain benchmarks, and four OOD benchmarks, SERPO improves HealthBench and ResearchQA by up to 20.63 and 20.31 points over the corresponding base models, raises the six-benchmark macro-average by up to 8.06 points, and supports OOD transfer and continued cross-benchmark evolution.

Human personality theories characterize traits not as isolated attributes captured by a single score, but as stable individual tendencies expressed through the interplay among persons, situations, and behaviors. Existing studies of personality-related behavior in LLMs have primarily focused on outputs elicited under personality conditioning, characterizing observable trait-related expressions while lacking mechanistic evidence for the existence of internal personality-related representations, their cross-situational expression, and how these representations shape specific behaviors. Building on Funder's personality triad framework, we adapt its three components for LLM analysis: Person as personality-related internal representations, Situation as contexts that afford trait-relevant responses, and Behavior as response patterns on broader social tasks. We introduce a framework for discovering, controlling, and validating trait-like representations in LLMs. First, using contrastive behavior pairs grounded in shared situations, we identify sparse internal features associated with opposing poles of personality traits through SAE decomposition. We validate their trait relevance through effects on behavior to situation, token-level activation patterns, and robustness to paraphrasing. Second, feature-level interventions induce bidirectional trait-related shifts across a separate, diverse set of situations while preserving response validity, demonstrating consistent expression across contexts. Third, applying the same interventions to social intelligence tasks reveals behavioral changes with benefit-tradeoff patterns consistent with findings from human personality research, providing behavioral-level validation beyond personality scores. Our findings provide evidence that LLMs contain controllable trait-like representations linking internal states, situational expression, and behavioral outcomes.

Despite the existence of exponentially many valid tokenizations for a given string, language models operate on a single canonical sequence deterministically produced by the tokenizer, leaving the broader tokenization space largely uncharacterized. In this paper, we investigate this overlooked space by studying the behavior of language models under non-canonical tokenizations across diverse languages. For English, prior work shows that models are largely invariant to alternative tokenizations that represent the same underlying string. We ask whether this invariance generalizes to other languages beyond English. We conduct a multilingual study across 27 languages spanning diverse scripts and evaluate LLM behavior under alternative tokenizations across six downstream tasks. We find that tokenization invariance does not generalize: model behavior varies substantially across languages with instruction-tuned models exhibiting an average relative performance drop of 23.7% for Llama-3.1-8B, 11.4% for Qwen3-8B, and 9.9% for Gemma-3-12B. The variation of tokenization invariance is systematic across languages. Languages that exhibit higher token fragmentation show significantly greater sensitivity to non-canonical tokenizations. Our study of tokenization robustness serves as a diagnostic of how tightly a model is coupled to its tokenizer. These results demonstrate that tokenization robustness is not a universal property of language models, but depends strongly on the language and its interaction with the tokenizer. We also show that LoRA fine-tuning with multi-tokenization training data provides an effective mitigation for tokenization sensitivity. Fine-tuning on English alone improves tokenization robustness across languages, while systematically sampling diverse non-canonical tokenizations achieves the strongest overall performance.

Large language models (LLMs) encode rich concept-like information, but represent it implicitly through distributed statistical associations rather than as explicit, structured, compositional concepts. Consequently, concept-level structure is typically found rather than designed: it is recovered after training through probing or dictionary learning, with no architectural guarantee of stability, compositionality, controllability, or alignment with human conceptual organization. We organize concept-aware interventions along two dimensions: whether concept structure is internally induced or externally grounded, and the stage of the pipeline where it is introduced. This taxonomy reveals three broad patterns: inference-time approaches remain comparatively underexplored, related ideas have developed largely in isolation across pipeline stages, and externally grounded methods span the entire pipeline despite often being described under different terminology. Together, these observations motivate moving beyond recovering concept-like structure from trained models toward designing LLMs with explicit conceptual representations.

Span-guided rewriting aims to preserve meaning by localizing edits to annotated harmful spans, but the same constraint can leave harmful intent insufficiently mitigated. We present a controlled exploratory comparison of span-guided and unguided detoxification on a mixed-source English evaluation set comprising manually curated inputs and HateXplain test items. We conduct a dense blinded human evaluation under a fixed single-generator setting. Human preferences reveal a trade-off rather than a uniformly superior rewriting strategy. Span-guided outputs are favored when localized editing preserves the original stance and avoids unnecessary modification, whereas unguided outputs are favored when broader rewriting achieves more complete mitigation. This contrast varies substantially across the study-defined strata: the two strategies are competitive in the strong stratum, while unguided rewriting is clearly preferred in the mild stratum. Rationale annotations trace this difference to complementary failure risks: residual harm after localized editing and over-modification after broader rewriting. We treat automatic evaluation as a diagnostic rather than a substitute for human judgment. Toxicity-similarity scalarizations, a multi-generator analysis, and two general-purpose LLM judges reproduce parts of the aggregate tendency but do not yield an analogous stratified contrast. These setting-specific findings do not establish a severity-based routing rule. Instead, they motivate evaluation protocols that assess mitigation sufficiency and meaning preservation separately and report both residual harm and over-modification alongside aggregate scores.

Large Language Model (LLM) agents are increasingly adopted in real-world security operations with access to host artifacts and command-line interfaces (CLIs), making it critical to thoroughly assess their security capabilities. However, existing cybersecurity benchmarks focus on pre-compromise settings where agents are placed in a clean and idealized environment before an attack occurs. This leaves the post-compromise setting underexplored. To address this gap, we introduce SecRespond, the first benchmark for evaluating LLM agents on the post-compromise incident-response workflow. Given a forensic disk snapshot of a compromised host together with the alerts, vulnerability scans, and baseline checks reported by a host security product, agents are required to produce forensic reports on intrusions, baseline risks, and vulnerability risks, together with a remediation plan. We instantiate this task across 10 cyber ranges, each constructed from a distinct compromised cloud host, spanning 4 entry-point types, 21 ATT&CK techniques, and 5 operating systems. We evaluate 23 frontier LLMs on the OpenCode agent harness. Experimental results show that although current agents can reliably uncover the problems exposed by alerts, they struggle to proactively investigate the disk for silent intrusions and to produce comprehensive, verified remediation plans, with no model achieving complete detection and remediation on any single range. This reveals a fundamental bottleneck in building agents for real-world incident response. The benchmark is publicly available at https://github.com/Alibaba-NLP/qqr/tree/main/data/secrespond.

The ability of large language models (LLMs) to process and generate text has introduced potential for applications in information extraction (IE). While it's debated whether LLMs outperform smaller fine-tuned models for classification tasks, their strong generalization capability makes them promising for domains with limited labeled data available for fine-tuning. This advantage is particularly relevant for the emerging application of the digital product passport (DPP), where the problem space is broad but domain-specific data remains scarce. Motivated by this use case, we apply generative IE to the product domain, explicitly addressing efficiency, generalizability, and data privacy constraints. We propose a two-step validation method that integrates a PLM block into the generative IE pipeline and thereby leverages LLMs' correction capability. We discover that such a validation task enhances LLM performance, particularly on the extraction of weakly expressed, low-salience entities that appear sparsely throughout the text. For certain entities, the performance of mid-size models can even reach levels comparable to larger models, and the improvement of first-step PLM predictions also enhance the final LLM output. Nevertheless, the effects on the smallest open-source LLMs (e.g., Llama-3.2 3B) is limited. Based on the findings, we develop a demo application for product information extraction that utilizes locally deployed LLMs, targeting further adaptations to real-world DPP use cases.

Code search in large-scale ecosystems is often hindered by the lexical gap between user queries and implementation details, alongside the trade-off between the low latency of traditional Information Retrieval (IR) and the precision of Deep Learning (DL). We present MediaWiki Code2Code Search, a neural retrieval system for semantic code-to-code discovery. By indexing 1.29 million structural entities (functions, types, and templates) across 2,500+ MediaWiki repositories, our system enables retrieval based on computational intent rather than surface tokens. We employ a split-build architecture, decoupling GPU-intensive offline indexing from a CPU-only serving layer; our FAISS IVF-PQ index occupies 168.6 MB: a 96.6% reduction compared to a flat float32 baseline, and achieves a median query latency of 1.85 seconds on commodity hardware, satisfying the 6 GiB RAM constraint of Wikimedia Toolforge. Our evaluation across a 27-query benchmark demonstrates superior performance over the BM25 baseline, achieving a P@10 of 0.87 compared to 0.64 (0.52 versus 0.34 for strict matching). Gains are most pronounced in name-obfuscated tasks where lexical methods fail. The system is available at https://code2codesearch.toolforge.org under the Apache 2.0 licence and provides an open RESTful API.

When it comes to generating vector representations of words, current language models are achieving high-quality results. However, what is not known is the extent to which knowledge about semantic relations is represented in the geometry of the semantic spaces created in this way. In order to answer this question, we study the relation geometry of such semantic spaces from three perspectives. We first examine whether words standing in a particular relation to a target word~(called relata) occupy the same region in semantic space, and whether the regions corresponding to different relations are distinct from each other. We then verify to what extent semantic spaces reflect certain well-known properties of relations, such as symmetry, asymmetry, and transitivity. Finally, we consider which information about the target words and relata is more important for relation geometry: their surface forms, or their contexts. We conduct experiments on six semantic relations using causal, masked, and diffusion language models. The results show that relata in asymmetric relations relatively clearly occupy a distinct region in semantic space. Asymmetric relations' properties are only moderately well encoded in the semantic space, yet better than those of symmetric ones. Furthermore, when considering the question which information source has the strongest impact on results amongst the models we evaluated, we find that lexical information tends to be more important for the causal language model, whereas contextual information is more important for the masked and diffusion language models. Our results empirically show that relation geometry is not equally well-represented for all relations in semantic space, suggesting that there is a difference in how well semantic relations might be learned from distributional information alone.

Recent advances in AI agents have increasingly internalized native capabilities into their underlying foundation models, giving rise to multimodal foundation models and large reasoning models. However, agent memory is still primarily implemented through external modules, leaving the native memory capability largely unexplored. In this paper, we take a first step toward this direction by introducing memory foundation models, which empower foundation models with native memory capabilities. We formalize native memory from two perspectives: a persistent and dynamically evolving memory state within the backbone, and native memory procedures that autonomously store and utilize information through model computation. We show that native memory offers advantages in architecture, end-to-end optimization, and efficiency. Based on this formulation, we propose Metis, the first prototype of memory foundation models. Metis introduces a new architecture that equips a foundation model with a native memory state, allowing historical information to be compressed into the model and accessed through memory attention. We construct large-scale memory-specific training data and introduce multiple optimization objectives to acquire these native memory procedures through mid-training. The online memory maintenance of Metis is gradient-free, and the memory update requires only a forward pass. At inference time, all learned model weights remain frozen, while the native memory states are autonomously transformed through standard forward computation. Through extensive experiments, we show that Metis exhibits native memory capabilities and further provide a detailed analysis of its strengths, limitations, and behaviors. To facilitate future research on memory foundation models, we release our project and model checkpoints.

State-of-the-art intracortical brain-to-text systems pair a neural-sequence phone decoder with an external language model. Two design axes remain underexplored: whether selective state-space models (Mamba) improve on recurrent decoders, and how the output target (phonetic vs.\ character) interacts with that choice. On the public Brain-to-Text '25 benchmark, we study a controlled 2x2 grid (GRU vs.\ hybrid Mamba decoder; phonetic vs.\ character targets) trained with a CTC objective under one reproducible protocol. The recurrent baseline remains strongest: the best phonetic GRU reaches 12.62% PER and 21.19% WER, while the best textual GRU after LM rescoring reaches 13.39% CER and 26.28% WER. The Mamba hybrid is competitive but does not surpass it. Ablations isolate architectural contributions, and error analysis shows representation-dependent failures: articulatory-like phoneme confusions vs.\ lexical and word-boundary errors.

Emotion Recognition in Conversation (ERC) plays a key role in developing emotionally responsive conversational systems, and contextual modeling is central to this task. Several studies have developed contextual models that capture dialogue history, speaker interactions, and emotion dynamics to support emotion prediction. However, the specific role of affective atmosphere, a latent, conversation-level emotional tendency shaped by conversational content and speaker interactions, remains insufficiently understood. This paper investigates what role atmosphere plays in conversational emotion understanding. We first propose AtmosERC, a graph-based atmosphere extractor that utilizes conversational context and speaker interactions to produce a compact representation of affective atmosphere. We then evaluate two downstream uses of the extracted atmosphere: initializing lightweight ERC model and providing emotion cues for LLM-based ERC. Experiments on three widely used ERC benchmarks show that the estimated atmosphere improves emotion prediction across both applications and supports more reliable predictions when local emotions deviate from the overall affective tendency.

The Mpox outbreak remains a serious public health issue, with the WHO (World Health Organization) reporting increasing cases in some regions. Research on Mpox is vital for several reasons, including vaccine development, diagnostic improvement, viral evolution studies, and preventing future outbreaks. However, the large amount of research being published makes it difficult to organize and analyze information efficiently. This study focuses on using multilabel classification to categorize 14590 Mpox research articles into key topics such as outbreaks, vaccination, and epidemiology. Among the different AI models tested, BERT performed the best, achieving 97.05% accuracy, 97.67% micro F1 score, and 96.46% macro F1 score. To better understand how the model makes decisions, SHAP was used to analyze significant word features and patterns. The results show that BERT can help automate the classification of Mpox research, making it easier for researchers, policymakers, and healthcare workers to quickly find relevant information, saving time and improving public health efforts.

The rapid growth of scholarly literature has made identifying relevant publications increasingly difficult, and conventional search systems still depend heavily on manually formulated queries and effortful manual inspection. Generative large language models (LLMs) offer a more flexible alternative, supporting literature retrieval and the screening of candidate studies against eligibility criteria. This chapter surveys 34 peer-reviewed papers applying generative LLMs to these two tasks, identified via a Boolean search over the OpenAIRE Graph (1,589 records screened to 34 inclusions). Reviewed studies are characterised by LLMs employed, model access and adaptation, prompting and architectural techniques, ground-truth sources, and evaluation metrics.

Post-training alignment is often shallow, eroding under fine-tuning. It remains untested as to whether constitutional midtraining interventions can produce durable alignment when cleanly isolated from post-training. We build a 394M-token constitutional corpus from Anthropic's Constitution and apply constitutional midtraining at 120B scale, where principled, values-based content is inserted into midtraining. A 2x2 design (curriculum ordering x deliberative reasoning) was used to produce four constitutionally midtrained conditions, plus a control, which were evaluated on self-generated and established benchmarks including alignment under pressure, value conflict resolution, blackmail, and emergent misalignment. All models were evaluated across three stages: post-midtraining, post-SFT, and post-benign fine-tuning. Constitutionally midtrained models outperformed the control on alignment generalization and durability, notably on blackmail: SFT instilled a blackmail propensity in all models, but constitutional midtraining blunted it, with the advantage surviving benign fine-tuning (-17.5pp). This durability did not extend to settings that required active resistance to in-context pressure or conflict, where the advantage attenuates after SFT. The presence of constitutional content at midtraining also mattered more than its structure, and constitutional midtraining incurred no capability cost, on average, at any stage (MMLU, ARC-Easy, piqa, GSM8K). A modest amount of constitutional content at midtraining could therefore yield broad, persistent alignment gains, offering a cheap, complementary addition to SFT-centered pipelines. Code, data, and models are available.