Papers

Resource constrained single-board computers including Raspberry Pi, NVIDIA Jetson Nano, Arduino UNO Q, Orange Pi, and LattePanda motivate on-device small language model (SLM) agents that reduce cloud dependence, improve data locality, and tolerate intermittent connectivity. Model Context Protocol (MCP)-style tool invocation demands more than fluent generation: an agent must emit machine-readable JSON, select the correct tool, supply all required arguments, and avoid unintended actions. We establish a platform-agnostic CPU baseline by evaluating five open-weight models below two billion parameters Phi-1.5, Pythia-1.4B, TinyLlama-1.1B-Chat, Qwen2.5-0.5B, and Qwen2.5-1.5B on 100 prompts spanning weather retrieval, web search, calculation, email composition, and task creation, under greedy decoding and nucleus sampling. A recovery parser strips Markdown fences, extracts brace-delimited substrings, and scores parseability, tool-name correctness, argument completeness, and value agreement. Under this criterion, Qwen2.5-1.5B achieves 75% (greedy) and 79% (sampling); Qwen2.5-0.5B achieves 72% (greedy) but drops to 32% under sampling. Phi-1.5 scores 0%; Pythia and TinyLlama reach at most 7%. A strict post-hoc audit finds only 5 of 1,000 raw responses directly parseable as JSON, exposing near-total dependence on output recovery. A CPU resource probe shows Qwen2.5-1.5B requires 7,960 MiB and 30.782 s mean latency; Qwen2.5-0.5B uses 3,637 MiB and 10.627 s, revealing a reliability-resource trade-off for edge deployment. These results do not cover the named boards directly or a full MCP implementation. Safe deployment requires schema validation, constrained generation, least-privilege execution, and human escalation for consequential actions.

Converting in-service reinforced-concrete (RC) building blueprints into simulation-ready models—structured frame representations that support deterministic FEM export and qualified-engineer review—underpins safety assessment and seismic retrofit, but the process remains manual. Direct prompting of a multimodal large language model (MLLM) over a scanned sheet is unreliable: outputs often violate engineering constraints on beam–column support, span count, or 3D continuity. We present BlueprintAgent (BPA), a constraint-triggered multimodal agent for simulation-ready frame extraction from scanned blueprints. BPA treats the MLLM as the primary reader and decision maker, with OCR and computer vision supplying localized evidence. Its central mechanism realizes engineering constraints as callable validators whose entity-level conflict reports trigger targeted MLLM revisits over the local region—an inference-time control distinct from fixed pipelines and free-form self-reflection. We evaluate BPA on 300 real scanned blueprint sheets from 20 anonymized RC frame projects, against five baselines and six ablations. BPA reaches a macro-averaged Beam F1 of 0.994, against 0.301 for single-MLLM zero-shot and 0.820 for a fixed pipeline; removing MLLM-led axis adjudication collapses Beam and Column F1 on complex multi-sheet projects. For dense technical drawings, engineering constraints are best deployed as triggers for entity-level targeted revisits rather than as post-hoc output filters.

Rotary position embedding (RoPE) uses each token's integer position to determine the rotation applied inside attention. This works well for local token order, but increasing context length creates a positional train-test mismatch: RoPE produces relative rotations at offsets not seen during training. Methods that rescale, interpolate, randomize, or bias positions specify how attention handles those offsets, but still derive positional information from a growing token counter. We instead divide a token stream into units, retain ordinary RoPE positions within each unit, and assign every completed unit an address computed from its content. Adding units then applies the same learned map to new content rather than extending a positional range or an identifier table. We prove that this construction preserves local RoPE exactly, leaves the attention comparison between two fixed tokens unchanged when other units are inserted or reordered, and does not create new relative rotations merely because more units are added. In a character-level Tiny Shakespeare diagnostic, all-token validation perplexity remains approximately constant from contexts of 256 to 4096 characters. A second diagnostic shows that content-based addressing can retrieve and use information from multiple serialized facts. These are controlled shallow experiments, not scale benchmarks, but they support a direct prescription: use position to address locally and content to address across units.

Robustness evaluation of large language models (LLMs) remains a critical challenge, particularly in assessing their sensitivity to perturbations in input data. In this work, we systematically evaluate LLM robustness across multiple dimensions, including word error rate, character repetition and duplication, modifications in choices, and variability in instruction following. To facilitate this evaluation, we construct a synthetic and augmented dataset encompassing a diverse set of LLM benchmarks, specifically targeting multiple-choice question (MCQ) datasets and instruction-following tasks. We conduct extensive experiments on LLMs of varying scales-small, medium, and large-as well as across base and instruction-tuned variants. Our analysis quantifies the variability in model responses under perturbed conditions and highlights discrepancies relative to baseline models. The findings provide insights into the stability of LLMs across different evaluation scenarios contributing to the development of more robust and reliable language models as well as robust evaluation methodologies.

Taxonomy induction aims to organize concept sets into coherent hierarchical structures. Recent LLM-based methods can induce taxonomies directly from flat term lists, avoiding the need for corpora, but degrade sharply as concept sets scale up. We argue that this degradation stems not only from context length limitations, but also from structural failures in hierarchical reasoning. To address this, we adopt a divide-and-merge paradigm that partitions concepts into smaller subsets, induces local taxonomies, and merges them into a global hierarchy. However, we identify two structural failure modes inherent to this paradigm: Structural Fragmentation, where partitioning weakens local hierarchical signals, and Parent Displacement, where locally plausible relations are misplaced in the global hierarchy. To address both, we propose SPARROW, a scalable taxonomy induction framework that combines structure-preserving spectral partitioning to retain hierarchical connectivity within each block, and constraint-guided incremental fusion that treats block-level relations as structural constraints rather than ground truth for global placement. Experiments on large-scale benchmarks show that SPARROW consistently achieves the strongest global structural quality across backbones. The code is available at https://github.com/rebeccazyr/SPARROW.

Dynamic sparse attention reduces long-context prefill cost by routing each query chunk to a small set of key chunks at every Transformer layer. The sparse attention kernel avoids most token interactions, but the router still rebuilds a chunk–chunk score matrix layer after layer, even when the selected routes change little. We introduce RouteRelay, a router-agnostic method that reuses only route metadata across depth while continuing to compute attention with the current layer's queries, keys, and values. Anchor layers perform full routing. Intermediate layers rescore the previous top-$k$ route and a compact sentinel set of near-miss and randomly probed chunks. A query row is rerouted only when a sentinel challenges its weakest selected chunk. We give a top-$k$ stability condition, a probabilistic bound on missed challengers, and a row-selective GPU execution design. In a reproducible empirical evaluation, RouteRelay retains at least 99.99% route recall while rerouting 25.0%, 55.4%, and 78.2% of rows under low, moderate, and high cross-layer drift, respectively. Across routing scales, RouteRelay retains 100.0% recall while evaluating 38.4–51.6% of full-routing score pairs as the key-chunk count grows from 128 to 1024. Its unfused CPU execution remains slower than dense matrix multiplication, exposing row compaction and ledger updates as the main kernel-engineering targets.

Demographic synthetic survey panels are often validated by matching aggregate answers to published surveys. We test what that certificate establishes across six multiselect batteries from four survey organisations in three countries. The headline analysis is restricted to three instruments whose synthetic cohort and human target share the stated population frame; three other batteries remain sensitivity analyses. The response contract dominates measured fidelity. In the aligned instruments, committed sets leave 66 of 128 model-battery option slots empty in panels of up to 500 respondents, versus 0 of 128 under per-option probability elicitation. Across eight uncapped model-instrument comparisons, probabilities reduce option-marginal MAE by 4.53 to 7.30 points. The capped instrument reverses on two models until the vectors are projected onto its stated maximum. These are measurement effects: human targets are realised check-all responses, whereas the vectors are latent inclusion propensities. Published marginal agreement also fails to discriminate respondent simulation from direct population estimation. On nine aligned model-battery pairs, a no-persona population-prevalence query averages 6.27 MAE versus 12.39 for committed panels and wins all nine comparisons. Constraint-aware probability vectors average 5.34 and beat the query on four of nine, so the baseline challenges the validation criterion rather than proving direct estimation uniformly best. On three unpublished demographic cells, neither approach beats reciting the national distribution. Population-marginal agreement is therefore evidence about an elicitation contract and an estimand obtainable without simulated respondents, not evidence of individual simulation.

Ethical evaluation of Large Language Models (LLMs) often characterizes model values as static and monolithic. In contrast, we argue that LLM value expression is better understood as a structured yet dynamic phenomenon. To investigate this, we introduce Conflict-driven Value Probing, a controlled framework that places LLMs in value conflicts and implements four types of interventions that perturb these conflicts to probe LLM value expression. Applying this framework to ten LLMs, we identify three recurring patterns. (1) Expression duality: models shift from broad idealistic orientations in abstract assessment toward more pragmatic priorities in concrete conflicts. (2) Functional steerability: models readily reconfigure their expressed value profiles toward task-defined value objectives. (3) Bounded plasticity: such reconfiguration is not without constraints, i.e. pressure induces a security- and goal-oriented priority shift while negative framing distinguishes protected values from those more amenable to redirection. Together, these findings characterize both the structure and dynamics of LLM value expression: context flexibly reconfigures expressed priorities, yet within behavioral boundaries. This behavioral account provides a foundation for understanding controllability, alignment, and safety in LLMs. Code and data are available at https://github.com/ZeroGen-Lab/CFProbe.

Large audio-language models (LALMs) can hallucinate audio objects, answering "yes" to absent sound events, thus undermining reliability in audio question answering. We propose Token-Adaptive Decoding (TAD), a training-free strategy for hallucination mitigation that grounds the initial yes/no decision by contrasting logits under real audio with a matched silent reference. TAD introduces a token-adaptive, confidence-guided gate that is decision-critical at the first decoding step and class-conditional on affirmative tokens, using the audio-silent margin to avoid overcorrection when evidence is weak or already sufficient. Experiments on AudioCaps-Hallucination show that, relative to Audio-Aware Decoding (AAD), a contrastive baseline with fixed contrast strength, TAD improves F1 for Qwen2 by 0.059 to 0.117 across Popular, Adversarial, and Random splits, and for Gemma by 0.025 to 0.064, while on Clotho-AQA it raises F1 from 0.810 to 0.816 on Qwen2 and remains comparable to AAD on Gemma.

Methods for streaming language models are often discussed alongside long-context and memory systems, although they solve different problems. An attention sink can stabilize autoregressive generation over an indefinitely long stream while the model remains unable to use content that has left its recent-token cache. We argue that this distinction should be explicit in system claims and evaluation. We introduce three horizons: the stability horizon, over which predictive behavior remains well behaved; the access horizon, over which past content can still causally affect the output; and the utility horizon, over which a task retains acceptable performance. We show constructively that the stability horizon can be infinite while the access and utility horizons are finite. We then propose ThreeH, an evaluation contract that measures all three horizons under a common state and compute budget. Applying the framework to attention-sink streaming clarifies its strength, constant-memory, stable generation, without treating anchor tokens as semantic memory. The framework exposes roles for cache policies, recurrent state, retrieval, and external memory. Experiments on 128K-token streams, delayed binding recall, and delayed decisions show that attention sinks preserve local modeling but not content beyond the active cache; recurrent and retrieval state extend the semantic horizon.

Post-hoc sparse attention accelerates long-context prefill by routing each query to a small set of token-level interactions. Hard selection, however, assigns zero probability to every omitted chunk: a routing miss cannot be recovered, and a fixed expansion budget spends the same work on easy and ambiguous queries. We introduce Coarse-to-fine Error-aware Dynamic Attention Routing (CEDAR), a coarse-to-fine method that keeps the language model frozen while preserving global coverage. Each semantic chunk contributes a cheap key–value summary to a residual attention path; chunks with high estimated approximation error are then expanded to exact token attention. Exact and summarized contributions are combined in a single softmax normalization, so refinement replaces, rather than duplicates, coarse evidence. We derive an output-error bound governed by within-chunk key/value dispersion and use it to allocate a variable refinement budget. A controlled clustered-attention study shows that residual summaries reduce reconstruction error by more than 98% relative to hard dropping at equal exact-chunk budgets. Experiments on long-context benchmarks demonstrate that CEDAR recovers most of the quality lost by hard sparse routing while maintaining approximately $3\times$ kernel speedup at 128K context.

Realizing full-duplex spoken dialogue requires large amounts of two-channel, one-speaker-per-channel conversational speech data. Although conversational text-to-speech (TTS) engines have been developed, they are not necessarily robust to two-party simultaneous phenomena such as backchannels, interruptions, and overlaps that occur while the interlocutor is speaking. In this work, aiming at conversational speech synthesis that reproduces human-like overlap, we propose KABURI-TTS. KABURI-TTS takes a per-speaker phoneme raster as input and renders the speech of the two speakers on separate channels, conditioned on the per-frame phonemes and the voice activity derived from them. Because the phoneme raster is supplied by a separate module, the proposed method enables controllable generation of one-speaker-per-channel, two-party spoken dialogue. A user evaluation shows that, compared with strong baselines, the proposed method attains higher naturalness at both the utterance and the interaction level. Furthermore, an analysis of voice activity confirms that the proposed method produces more overlap and more frequent turn-taking.

Background: Large language models (LLMs) show promise for extracting information from clinical free-text documents, but their outputs are often unstructured and lack traceability, complicating validation and adoption in clinical workflows. In this work we introduce SIFTING, an LLM-based framework designed to address these shortcomings. Methods: SIFTING combines the language comprehension capabilities of LLMs with segment-level processing and structured prompts with strict output control, linking findings to the source text to enable both accurate and transparent information extraction. To demonstrate its capabilities, we applied the framework to the task of extracting tumor T-stage information from 130 lung cancer radiology reports (SIFTING-T-stage). A compact 4-bit quantized version of the open-source LLM Llama-3.3-70B (35 GB) was used in a fully self-hosted setup, providing full control over data and model. Performance was evaluated against a reference standard created by four clinical experts and compared with a range of LLMs as used in a conventional single-prompt approach, using bootstrap resampling to estimate confidence intervals. Results: SIFTING-T-stage achieved an accuracy of 90% (95% CI: 84-95) against the reference standard. We found its performance to be comparable to even the largest state-of-the-art LLMs with reasoning capabilities and to be interchangeable with clinical experts (p < 0.001), while at the same time offering full traceability through source text references. Conclusion: SIFTING enables accurate, structured, and traceable information extraction from clinical free-text documents. It ensures data control, reproducibility, and verifiable outputs that can support clinical validation and workflow integration.

Reinforcement learning with verifiable rewards (RLVR) is sensitive to which problems a model trains on, yet existing selection criteria–difficulty filtering, hand-curation, reward-trajectory scoring–assess data value as an intrinsic property of problems, independent of the model that will learn from them. We introduce Circuit Reasoning Score (CRS), a selection signal derived from 46 reasoning-sensitive attention heads identified via contrastive ablation, computed in a single forward pass on the frozen base model without reward labels or rollouts. CRS runs against the intuitive hypothesis that stronger reasoning-circuit engagement produces better training data: on Qwen2.5-Math-7B, the lowest-engagement decile improves over random selection on three medium-difficulty benchmarks (GSM8K +2.0 pp, OlympiadBench +1.6 pp, Minerva +2.9 pp), while the highest-engagement decile gains less and is indistinguishable from the middle decile. The advantage has boundary conditions: on a domain-curated pool no selection method separates from the others; at 1.5B scale the useful direction differs; and the lowest-reward training condition produces the strongest downstream generalization. Within the Qwen2.5-Math settings tested, RLVR data selection appears regime-dependent rather than reducible to a static ranking of problem quality.

Diffusion Large Language Models (dLLMs) generate text via bidirectional iterative denoising, naturally supporting user-specified constraints anchored at arbitrary output positions, a paradigm known as In-place Prompting (IPP). We formalize this as the In-place Instruction Following (IIF) task and construct IIF-Bench, a hierarchical benchmark spanning literal, style, and discourse-function constraints, paired with a rubric-based local-global evaluation protocol. An inference-time attention-bias probe suggests that vanilla dLLMs often under-prioritize constraint spans during denoising. We then propose GRAFT, an IPP-oriented post-training framework combining constraint-aware SFT and preference optimization. On four representative dLLMs, GRAFT raises the average IIF score from 57.75 to 73.10 (+15.35 points), with absolute gains of 15.91 and 15.57 points on literal and discourse-function constraints, while preserving general generation ability.

Minimal-pair benchmarks such as BLiMP evaluate linguistic knowledge by testing whether language models (LMs) prefer acceptable sentences over minimally different unacceptable ones. However, these benchmarks largely ignore lexical frequency variation, despite lexical frequency being a pervasive and highly skewed property of natural language use. Consequently, existing evaluations do not test whether grammatical preferences remain stable when contrasts involve rare lexical items. We introduce FreqBLiMP, a frequency-controlled extension of BLiMP that regenerates all 67 paradigms under explicit Zipf-frequency regimes while preserving each minimal-pair's grammatical contrast. Evaluating multiple open-weight LLM families across scales, we find that decreasing lexical frequency produces a consistent, monotonic decrease in sentence likelihood, but only a modest reduction in overall contrastive acceptability accuracy. However, this aggregate stability masks substantial variation across linguistic phenomena, with LLMs remaining robust on overt morphosyntactic generalization while degrading on phenomena that require lemma-specific information.

Voice phishing (vishing) unfolds in real time; by the time a call has ended and post-hoc classification is possible, the harm has already been done. The more actionable question is which concrete harm (Information Gathering or Financial Exploitation) an ongoing call is tactically progressing toward. We present Vishing-Tactics-Bench, a benchmark grounded in Endsley's situation-awareness (SA) framework that recasts vishing defense from after-the-fact fraud classification to harm projection: predicting at each turn whether the call will reach either terminal harm. We adapt MITRE ATT&CK to vishing as a 6-tactic taxonomy (Vishing-Tactics) and label 35,340 scammer utterances across 5,645 synthetic Chinese calls. We define Exploitation Trajectory Forecasting, a survival-style protocol over the two terminal harms with three metrics: AP@k, C-index, and divergence error. Baselines ranging from a Markov heuristic to fine-tuned LLMs show that the tactical trajectory serves as an interpretable representation of the call's tactical state, supporting harm-specific forecasting, which can then be used for the downstream application of intervention selection; a stratified lead-time analysis at a tight false-alarm budget further identifies at what point in a call the trajectory signal yields early warning.

A transformer can make an attribute linearly decodable in its residual stream at a depth where that attribute does not yet influence the output. This gap between where information is readable and where it is used has been shown for attributes stated directly in the input. We ask whether it also holds for an attribute the model must infer gradually over a conversation, namely how expert its dialogue partner is. Using ExpertCollab, a corpus of multi-turn research-planning dialogues between model-played personas at four expertise levels, we find that partner expertise is most decodable in the early layers and falls to near chance before the midpoint of the network. Counterfactual patching shows that injecting the expertise difference at the layer of peak decodability barely changes a fixed late-layer readout, whereas the same difference injected past the midpoint propagates almost completely, a separation of more than an order of magnitude. A content-matched random control and a probe-free diagnostic place the transition at the same early layer, and a statically specified control attribute stays decodable throughout. An inferred relational attribute is therefore represented well before it becomes causally active, which bounds where any attempt to read out or steer partner-conditioned behavior must intervene. We use one model on a synthetic corpus as an initial demonstration.

This paper presents our system for CCL2026-Eval Task 5: Minor-Grain Breeding Information Extraction (MGBIE), which jointly extracts 12 entity types and 6 relation types from minor-grain breeding literature. We propose RAME (Retrieval-Augmented Multi-Prompt Ensemble), a training-free framework that elicits multiple LLM outputs under controlled diversity and aggregates them by majority voting to obtain high-confidence predictions. RAME combines (i) retrieval-augmented few-shot selection via a hybrid BM25-embedding retriever, (ii) a three-prompt ensemble (Strict, Relaxed, Balanced) spanning the precision to recall spectrum, and (iii) large-scale repeated sampling with majority voting to filter noisy predictions. Built on DeepSeek-V4-Flash, RAME achieves a Total Score of 0.499 (NER 0.730, RE 0.346) on the leaderboard, ranking 1st and surpassing the official Track-A baseline powered by GPT-5.5 (0.448), representing an 11.4% relative improvement. Code is available at https://github.com/king-wang123/CCL26-RAME.

Autoregressive language models generate one token per decoding step, limiting the useful output of each forward pass. Although diffusion models, insertion-based decoding, and multi-token prediction enable parallel generation, they either incur additional training-time token traffic or struggle to predict strongly dependent future tokens. We introduce the Line-Coupled Language Model (LCLM), an autoregressive model that advances multiple text lines together by predicting the next token for every active line while coupling the lines through shared causal context. LCLM interleaves line tokens into a single causal sequence and uses line-staggered rotary positions, retaining the standard next-token objective and causal attention. Controlled experiments show that cross-line targets are substantially less dependent than consecutive same-line targets, supporting lines as parallel generation units. With 881M parameters, LCLM produces an average of 2.94 content tokens per forward pass with a validation cross-entropy loss of 2.44, compared with 1.00 token per forward pass and a loss of 2.39 for the vanilla autoregressive baseline. Most notably, even when LCLM generates 16 tokens per forward pass, its loss is only 0.09 higher than that of the vanilla autoregressive baseline (2.34 vs. 2.25).

The ability to generate counterarguments is important for critical thinking and balanced discourse, yet existing approaches typically produce only a single counterargument, failing to capture the diversity and persuasiveness required in real-world debates. To address this limitation, we propose Persona-guided Tree-based Counterargument Generation (PTCG), a framework that combines Tree-of-Thoughts-inspired step-wise generation and pruning with speaker persona selection. By estimating the author's persona from the original argument and incorporating speaker personas representing distinct perspectives, PTCG operationalizes perspective-taking and enables the generation of diverse counterarguments. Results from LLM-as-a-Judge, classifier-based assessment, and human evaluations indicate that PTCG shows consistent improvements in both the diversity and persuasiveness of counterarguments compared to baseline methods.

Prompt-based interventions: system prompts, personas, role instructions, reliably reshape what a language model says, but it is unclear which layer they reach. Do they reconfigure internal structure, or only modulate the output channel? We use persona conditioning as a controlled probe, measuring its effects along a depth axis from self-report, through open-ended generation, to word-level parametric association, across three instruction-tuned models. We find a graded dissociation. Personas are legible but not structural: models follow single-trait instructions yet fail to reproduce human inter-trait covariance. The dissociation deepens with depth: personas hold or amplify closed-form QA bias, shift absolute tone while leaving between-group disparity unchanged, and barely perturb an already saturated associative baseline. Prompt-based steering thus operates in the output channel and has a structural reach limit that surface manipulability can mask.

Large language models (LLMs) are often post-trained on pre-collected reasoning trajectories to improve their reasoning capability. Such trajectories tend to be long due to complex, interwoven paths, which often include detours on the path toward the answer. However, it has been underexplored whether LLMs indeed benefit from learning complete trajectories in post-training, such as supervised fine-tuning (SFT). Starting from our pilot study, we find that full trajectories provide only limited benefit, while partial trajectories are effective even under heavy truncation. We analyze redundancy in reasoning trajectories through attention-based analyses and controlled token-removal studies, both of which show that intermediate tokens contribute minimally to final reasoning quality. This suggests that avoiding redundant information may allow LLMs to internally infer coherent alternatives by inferring missing steps from their internal knowledge, given known trajectory endpoints. Furthermore, we show that training LLMs using endpoints leads to consistent changes in reasoning behavior, and that it also benefits post-training methods based on reinforcement learning or on-policy distillation, highlighting the need to revisit complete reasoning traces. Code is available at https://github.com/naver-ai/revisiting-trace.

LLM assistants often produce more answers than humans can review before users see them. Most evaluations ask whether an answer is wrong, unsupported, or low-confidence. Bounded review budgets instead ask which answers should be checked first under a fixed review budget. Risk alone is not enough: a high-risk answer may be hard to repair, while a moderately risky answer may be directly correctable from available evidence. For generated-answer evaluation, we model review prioritization as exposure reduction, where review value combines estimated wrongness, intervention affordance, impact, and cost. We evaluate review queues with Wrong-Answer Exposure Ratio (WAER), the fraction of wrong answers left unreviewed, and post-repair residual exposure (PRRE), the fraction still exposed after deterministic benchmark-supported repairs. PRRE uses repairability rules that do not numerically reuse the affordance scores used for ranking. On a 720-item TAT-QA/SciFact stress benchmark, review-value ranking keeps answer-level WAER nearly unchanged at 20% budget (0.605 vs. 0.600) but lowers PRRE from 0.881 to 0.716. These results show that trustworthy LLM evaluation should measure not only error detection, but also how limited review capacity reduces exposed wrong answers.

Retrieval-augmented generation (RAG) enables large language models (LLMs) to answer questions by accessing external knowledge and has been widely adopted for long-term conversational memory question answering. However, existing methods suffer from two key challenges: (1) fragmented evidence scattered across temporally distant sessions, and (2) noisy content within retrieved sessions that triggers the lost-in-the-middle effect. To address these challenges, we propose MemLoc, a unified Retrieve-Localize-Generate framework for long-term conversational memory QA. For retrieval, MemLoc decomposes each session into multi-granularity memory units and performs query routing via an inner-memory graph with entropy-based granularity selection. It further models cross-session semantic and temporal dependencies through a cross-memory graph, enabling coarse-to-fine retrieval of top-K relevant memory candidates. For localization, we introduce a reasoning-based evidence locator trained with Self-reflective Hint Policy Optimization (SHPO), which performs progressive refinement by extracting query-relevant fragments within memory units to suppress noise and reranking across candidates to remove redundancy, producing a compact evidence set with lightweight location IDs. For generation, these IDs act as precise grounding signals that guide the LLM to the correct memory positions, mitigating the lost-in-the-middle effect while preserving original contextual integrity. Extensive experiments on four benchmarks demonstrate that MemLoc achieves state-of-the-art retrieval accuracy and response quality while maintaining efficiency. Our code is available at: https://github.com/Nikol-coder/MemLoc.

Voice phishing detection faces three critical challenges: real criminal recordings are unavailable due to privacy constraints; when available, only a handful of samples exist, insufficient for fine-tuning; and lightweight acoustic-only detection is needed as an alternative to large self-supervised models. We compare domain-invariant prosodic features and self-supervised representations (HuBERT, wav2vec2.0) through cross-domain evaluation-training on scenario-based actor recordings and testing on authentic criminal calls. Domain-invariant prosodic features achieve 69.5% F1 zero-shot and 71.0% with 5-shot learning. HuBERT achieves highest performance (94.2% F1, 5-shot), while wav2vec2.0 exhibits a precision-oriented detection profile (90.2% F1 with 99.4% precision, 5-shot). These findings reveal fundamental trade-offs: domain-invariant features enable zero-shot deployment when no real data exists, while SSL methods achieve higher performance but require real samples and compute.

LLM hidden states are ordinary vectors, but the distances among those vectors may still show hierarchical structure. To our knowledge, this paper is the first systematic study of whether prompt-token hidden states in contemporary LLMs exhibit Gromov Hyperbolicity (GH), a distance-based measure of tree-likeness. Using 818,904 sample-layer measurements from ten open-weight models across MATH500, HumanEval, WinoGrande, and TruthfulQA, we build a GH map over four axes: parameter scale, layer depth, model family, and input domain. The clearest pattern is depth, not scale: middle layers usually form a high-relative-hyperbolicity plateau, while final layers often become substantially more tree-like. Scale effects are weak and non-monotonic, matched 7/8B model families differ strongly, and domains interact with model specialization. These findings make GH useful as a practical diagnostic: it shows where hierarchical distance structure appears, how specialization changes it, and which model-layer-domain comparisons deserve closer analysis.

We identify the Flow Moment, a reasoning pattern characterized by sustained, process-confirming verbalizations such as I'm doing, in contrast to the revision- and backtracking-oriented Aha Moment. We refer to their corresponding linguistic expressions as Flow Markers and Aha Markers, respectively. Based on this observation, we construct Flow-CoT by rewriting the discourse markers of original reasoning traces while preserving their underlying reasoning content, and use it as auxiliary supervision for on-policy self-distillation (OPSD). We further propose Aha-Flow Distillation (AFD), a dual-mode extension of OPSD that pairs different forms of privileged information with corresponding reasoning instructions. The Aha branch retains concise solution-based supervision, while the Flow branch introduces rewritten Flow-CoT under a direct and confident reasoning instruction. At inference time, the model uses only the standard reflective instruction, so Flow-style reasoning serves purely as a training signal. Experiments on AIME25 and HMMT25 show consistent improvements across Qwen3-8B and Qwen3-4B: AFD improves Avg@12 from 60.8 to 61.3 on Qwen3-8B and from 57.5 to 58.6 on Qwen3-4B over our reproduced OPSD baselines. Controlled ablations further show that, with the same Flow-CoT/Aha-CoT composition, dual-mode training improves Avg@12 from 59.5 to 60.1, indicating that the benefit comes not only from introducing heterogeneous reasoning supervision, but also from how it is organized during self-distillation. The code is available at https://github.com/Wang-Xiaodong1899/Aha-Flow-Distillation.

Large language models (LLMs) increasingly generate Markdown that is consumed by renderers, agents, code extractors, and structured downstream pipelines. Yet existing evaluations often conflate content quality with format adherence, leaving Markdown boundary failures under-measured. We introduce LatentMD, a benchmark and evaluation protocol for diagnosing CommonMark-level fence-boundary failures in LLM-generated Markdown. LatentMD separates content correctness from boundary correctness, enabling detection of outputs that are content-correct but boundary-broken. The benchmark contains 4,179 prompts and a CLI for scoring arbitrary model outputs. Across 9 LLMs and roughly 37,600 generations, we find that Markdown boundary failures are widespread: 38.0% of valid main-grid outputs are content-correct but boundary-broken, with substantial boundary breakage under unspecified prompts and in a small human-authored validation set. Ablations show that failures are driven primarily by same-family symmetric-delimiter collisions rather than nesting alone, are only partially mitigated by prompt hints, and generalize to Python triple-quote docstrings while JSON remains robust as an asymmetric-delimiter control. LatentMD provides a reproducible diagnostic target for parser-sensitive LLM evaluation.

Large language models achieve strong performance across diverse tasks, but deployment remains costly because of memory, latency, and energy demands. Structured pruning reduces these costs by removing architectural components, yet its recovery stage is often limited by a mismatch between the recovery module's representational capacity and the complexity of the removed knowledge. We call this bottleneck the capacity-knowledge asymmetry and propose OverRep, an Overcomplete Reparameterization framework for structured LLM pruning. Following the principle of "train overcomplete, deploy compact", OverRep temporarily overparameterizes the recovery module during training to absorb complex knowledge distilled from the original model. After recovery, the overcomplete re-parameterization is algebraically merged into a mathematically equivalent compact module, preserving the pruned model's inference-time architecture and computational cost. OverRep further introduces an annealed activation that enables nonlinear training dynamics while converging to a linear regime for exact algebraic merging. Across three backbone families, OverRep improves retained reasoning performance over strong recovery baselines by up to 5.5 and 8.4 points at 25% and 50% pruning, respectively, while keeping memory usage and TFLOPs comparable to existing recovery methods. Our code is available at https://github.com/mmai-laboratory/OverRep.