Papers

Filipino graduates face a persistent disconnect between educational preparation and labor market outcomes, where starting salary is a key signal of entry-level valuation. Current Philippine research is dominated by descriptive tracer studies that document employment rates but do not explain the determinants of pay. We address this gap using a crowd-sourced survey dataset of graduate responses whose noisy, self-reported nature makes it a challenging prediction target. Applying machine learning to this problem, we identify job role and industry as the dominant determinants of starting salary, significantly outweighing institutional prestige. The strength of this finding is its central contribution: it is corroborated by three independent lines of evidence, namely SHAP attributions, the heavy reliance of the best ensemble on occupational text, and a Natural Language Inference reformulation. These results suggest that career guidance and policy should prioritize sector-specific skills over institutional brand.

Multiple-choice benchmarks fix the questions and the correct answers, but not the harness: the order of the options, the wording of the prompt, and whether a language model's answer is read from generated text or from per-option likelihoods. Work on this harness sensitivity reports it as aggregate score variance, leaving unexamined which items the variance falls on and whether they are the items that separate one model from the next. We treat the evaluation harness of large language models (LLMs) as an independent variable and resolve its effect to single items. We introduce the fragility grid: 12 open-weight instruction-tuned LLMs from 4 families answer the same 3{,}679 items from 4 benchmarks (ARC, HellaSwag, MMLU, TruthfulQA) under 26 equally defensible harness configurations, recording one correctness bit for every model, item, and configuration. The comparison is matched, since the items, the weights, and the greedy decoding stay fixed while only the harness varies. Under the grid a model's score is a band rather than a point: gemma4-31b scores between 31 and 89 percent depending only on the harness. Three results follow. On the items that two adjacent models both answer stably the pair is tied, and config-fragile items carry 95.7 percent of a pair's gap on average. Four of the 12 models reach rank one under some configuration, so the harness selects the winner. Item discrimination, the property that benchmark-compression methods maximize, correlates with fragility at 0.28 (95 percent CI 0.25 to 0.30), so compression keeps the fragile items rather than removing them. The scoring choice, not the option order that protocols usually fix, is the load-bearing axis. We release the per-item records and the analysis script, from which every number regenerates on a CPU in seconds, and we position the fragility grid as a check a leaderboard can run before it reports an order.

Large Language Models (LLMs) often require carefully crafted prompts to unlock their full potential, which can be a barrier for non-expert users. This work addresses the challenge by introducing a Task-Aware Prompt Rewriter (TAPR), a model that reformulates user prompts into task-optimized prompts with the explicit goal of improving downstream LLM performance. We train TAPR using reinforcement learning with Group Relative Policy Optimization (GRPO), where rewards are derived from LLM-as-judge evaluations of both the reformulated prompt and the corresponding task output. Experimental results on diverse tasks, such as question answering, summarization, and arithmetic reasoning, show that our method yields consistent gains over base models in prompt rewriting ability. Fine-tuning Phi-4-mini-instruct (as the base model for TAPR) produces prompts that contain clearer and more instructive language, leading to higher accuracy on established benchmarks such as Natural Questions and GSM8K. Our code is available at: https://github.com/OliverSavolainen/task-specific-prompt-rewriter

Wafer fabrication exhibits unique characteristics, including reentrant process flows, variable bottlenecks, and highly variable process conditions. In order to identify the most severe bottleneck at each moment in time for semiconductor wafer fabrication, this research presents the Dynamic Multi-Criteria Bottleneck Severity Index (DMBSI), a new, data-driven methodology for analysing multiple diagnostic signals of cycle time to changes in process parameters, and the impact of reworks on cycle time in order to generate an interpretable, unified measure of bottleneck severity. The experimental validation of DMBSI was conducted using Manufacturing Execution System (MES) logs collected from 22 wafer production lots at a commercial 200 mm wafer fabrication line operated by Seagate Technology. Using 5-fold cross-validation, the GA-optimised DMBSI achieves a Pearson correlation of r = 0.80 with observed cycle-time contributions, representing an 8.1% improvement over the expert heuristic baseline (r = 0.74) and substantially outperforming the Theory of Constraints (TOC; r = 0.60) and Value Stream Mapping (VSM; r = -0.30). Furthermore, the unique time-windowed component of DMBSI enabled the identification of temporal bottleneck migration patterns, which shifted from the dominant constraints associated with dielectric deposition steps in the early production windows to those associated with over- and under-inspection in the later production windows. The integrated what-if counterfactual analysis demonstrated that a 50% reduction in waiting time at the top-ranked bottleneck step would reduce the mean cycle time by 7.2%, with the top five bottleneck steps offering a combined potential reduction of approximately 19%.

Accurate prediction of crystal properties remains a key challenge in computational materials science. While graph neural networks (GNNs) such as CGCNN, MEGNet, ALIGNN, and SchNet have shown strong performance, they primarily represent crystals at the atomic level and implicitly learn local chemical environments through message passing. However, many material properties are governed by coordination polyhedra, the fundamental structural units formed by atoms and their neighboring atoms. To address this limitation, we propose the Coordination Polyhedron Graph Network (CPGN), a multi-scale GNN that jointly learns atomic, bond, and coordination-polyhedron representations. CPGN constructs three coupled graphs: an atom graph encoding elemental and bonding information, a line graph capturing angular interactions, and a coordination polyhedron graph describing Voronoi-derived local environments through corner-, edge-, and face-sharing relationships. Physically meaningful geometric descriptors are incorporated for each polyhedron, while an interleaved message-passing mechanism with bidirectional cross-attention enables effective information exchange across structural levels. Extensive evaluations on the Materials Project, JARVIS-DFT, and QM9 benchmark datasets demonstrate that CPGN outperforms existing state-of-the-art GNN models. It achieves a formation-energy MAE of 0.060 eV/atom and a band-gap MAE of 0.292 eV on the Materials Project, while providing competitive multi-property prediction on JARVIS-DFT and superior HOMO prediction on QM9. The results highlight that explicit modeling of coordination polyhedra improves crystal representation learning and enables accurate, physically interpretable prediction of material properties.

Digital mental health interventions (DMHIs) offer scalable support, but ensuring they accurately detect users' intent during volatile situations can be challenging. Pure parametric Large Language models (LLMs) do not contain specific safety critical architecture, and can miss critical cues, or hallucinate, undermining reliability. Retrieval Augmented Generation (RAG), which supplements an LLM with retrieved context, could enhance intent detection during volatile situations. Commercially available DMHIs typically combine multiple independent safety layers like rule-based filters, symbolic escalation protocols, and neural classification. The incremental contribution of any single layer, however, remains unquantified. This paper evaluates six LLM models within a DMHI called Wysa, via a controlled comparison of RAG-enabled versus RAG-disabled modes. Anonymized real and synthetic user-chatbot exchanges were annotated by a qualified clinical team against multi-class intent categories (e.g. self-harm, abuse, panic). The study computed classification accuracy, recall, precision and F1 scores against ground truth labels and tested differences for statistical significance. Performance was also examined by risk category and inter-model agreement. While RAG caused a rise in false alarms, the trade-off is consistent with safety-critical design principles that prioritize sensitivity, where flagged cases are routed to additional review rather than acted on directly. Overall, these findings support RAG as a promising approach to improve the accuracy, consistency and safety of LLM-driven DMHIs. Keywords: Digital Mental Health Intervention, Large Language Model, Retrieval Augmented Generation, Accuracy, Recall, Precision

High-resolution video diffusion models built on Diffusion Transformers (DiTs) deliver strong fidelity but quickly exhaust the memory budget of a single workstation. A 100 billion-plus parameter DiT easily requires over a terabyte of persistent state, while naive spatiotemporal self-attention grows quadratically in sequence length. These two walls – parameter memory and activation memory – prevent researchers from adapting massive generative models without large GPU clusters. We revisit this problem from a systems perspective and introduce MegaSlide-DiT, a prototype that demonstrates how a pre-trained 105B DiT can be adapted on a single H200 GPU with 1.5 TB of host RAM. Our key insight is that the GPU need not own the model state: all persistent weights, master weights and optimizer moments remain in host memory, while only transient shards are streamed to the GPU on demand. Simultaneously, we replace quadratic global attention with 3D Deformable Slide Attention (3D-DSA), a motion-adaptive local attention operator that reduces both memory and computational complexity to linear in the sequence length. We report detailed memory accounting, execution traces and evaluation results to substantiate our design. MegaSlide-DiT does not claim to train a 105B model from scratch on a single GPU, nor does it magically solve bandwidth limits; rather, it offers a pragmatic path for full-parameter adaptation of massive video diffusion models on high-end workstations.

Large Language Models (LLMs) are capable of generalizing human language for the completion of never-before-seen tasks, leading to widespread deployment. While this automation provides clear utility, completing these tasks often requires the insertion of Personally Identifiable Information (PII), strings of information that uniquely identify some individual, raising privacy concerns. However, ethics has prevented the curation of a public, authentic dataset of PII. Without an appropriate dataset, it is difficult to quantify privacy risks. Thus, we introduce the PANOPTICON pipeline and dataset. The dataset, generated by Meta's Llama-3.1-8B-Instruct model, contains 67, 718 prompts, intended for the models context window, containing PII spans derived from 9,674 publicly available synthetic user profiles. We measure lexical diversity and S-BERT diversity of the created dataset to evaluate realism. Finally, we present a case study showcasing the utility of PANOPTICON data for understanding Prompt Inversion Attacks (PIAs). PANOPTICON thus emerges as the first benchmark dataset for studying PIAs over private corpora, providing a foundation for future LLM privacy research.

Attention collapse in autoregressive language models – manifested as repetitive token loops where the model becomes trapped in self-reinforcing attractors – is a persistent pathology that existing decoding-time heuristics fail to address at its root cause. We present a principled framework that penalises or compensates anomalous confidence arising from collapsed generation patterns, by comparing a token's observed frequency against its corpus prior through an adjacent-conditional probability construction. The resulting self-normalising penalty ratio $R=f(m,n,p)/f(np,n,p)$ requires no ad hoc standardisation and admits a closed-form logit offset with zero approximation error. The correction is isolated from the loss gradient and accumulated into a frozen output-layer bias via exponential moving average, enabling deployment as a repair mechanism for models that have already collapsed without requiring intrusive modifications to standard training pipelines. Experimental validation on a 1.5B-parameter model demonstrates that the frozen-bias mechanism can rescue a model already trapped in a collapsed attractor, reducing 2-gram repetition from 0.073 to near 0 while preserving generation quality.

Large language models (LLMs) are increasingly used for emotional support despite lacking mechanisms to safely govern evolving mental health risk. Existing safety approaches primarily detect risk but rarely shape how models respond as conversational risk unfolds. We developed a model-agnostic safety governance architecture that combines contextual risk detection, reasoning-based verification, and protocol-guided response generation for multi-turn mental health interactions. Synthetic conversations grounded in real-world mental health narratives were used to evaluate the architecture's performance, tested with GPT-5-chat and Qwen3.5-27B, achieving high risk detection performance (specificity: 0.85 (95%CI: 0.78;0.91), sensitivity: 0.92 (95%CI: 0.88;0.95)) and increasing clinician-preferred escalation responses by 25.6–59.2pp while preserving rapport and connection. Performance remained stable across conversation length and generalized across both proprietary and open-source models. These findings demonstrate that clinically-grounded safety governance can extend beyond risk detection to improve how LLMs manage evolving mental health risk, providing a scalable framework for safer deployment across models.

Urban traffic congestion significantly increases fuel consumption, greenhouse gas emissions, and commuter delays, resulting in substantial economic losses and environmental harm in modern cities. Traditional traffic signal control strategies such as fixed-time scheduling, actuated control, and reinforcement learning (RL)-based methods, offer different degrees of adaptability; however, RL-based methods can require extensive retraining, careful reward design, and substantial simulation data when transferred across networks or demand regimes. To address these challenges, we propose HiLLTS, an LLM-guided traffic signal control framework that employs a hierarchical three-layer architecture consisting of a central coordination agent, a district layer and multiple cluster-level intersection agents. Experimental results demonstrate consistent improvements in both congestion and environmental performance. Compared with the strongest non-LLM baseline in each scenario, HiLLTS reduces average waiting time by 36.73% under the low-congestion scenario and 14.71% under the high-congestion scenario, while reducing average CO2 emissions by 7.87% and 8.57%, respectively. Larger gains are observed against weaker baselines: under low congestion, HiLLTS achieves reductions of up to 18.00% in emissions and 62.07% in waiting time relative to Fixed-Time control; under high congestion, reductions of up to 28.89% in emissions and 40.36% in waiting time are observed relative to Max Pressure. The ablation study further validates the contribution of LLM-guided coordination over rule-based control

Long-term memory enables LLM agents to leverage past interactions, but dialogue histories quickly exceed the context window, forcing agents to retrieve relevant subsets at query time. Because useful evidence is sparse and scattered across verbose conversations, retrieval faces a fundamental tension: broadening recall improves coverage but floods downstream reasoning with noise, while compressing memories at write time eases retrieval but irreversibly discards details that future queries may need. We introduce LazyMem, which resolves this tension by deferring all memory construction to query time. Given a retrieved candidate pool, a lightweight model processes it in overlapping parallel windows, selectively retaining and compressing only query-relevant content. The model is trained with supervised fine-tuning followed by reinforcement learning, using a reward that jointly encourages the identification of relevant messages and the generation of compressions that are faithful to the source and useful for answering the query. On LongMemEval, LazyMem-4B achieves an LLM-judge accuracy of 0.85, outperforming the strongest non-oracle baseline while using only 213 answer-context memory tokens, 21.0 times fewer than the baseline. It further generalizes to LoCoMo without target-domain training and reduces mean latency relative to the prior query-time baseline. Code is available at https://github.com/allacnobug/LazyMem.

Graphical user interface (GUI) agents are systems powered by large multimodal models (LMMs). They perceive screen state and execute user instructions through GUI actions such as clicking, typing, and scrolling on desktops and mobile devices. However, current agents scale poorly to long-horizon tasks: actions incur costly LMM inferences, and performance degrades as context grows. Humans divide such workloads among collaborators who complete sub-tasks in parallel. Yet parallel coordination among GUI agents has received little attention. To close this gap, we introduce ParaGUIBench, to our knowledge, the first benchmark dedicated to parallel execution and coordination of multiple GUI agents on separate desktop instances. It consists of three components: a multi-device Docker infrastructure with a shared file system; a dataset of 233 tasks spanning six task categories; and an evaluation system with efficiency metrics, including step reduction ratio and token cost. We further introduce ParaGUI, a planner-worker agent that decomposes GUI tasks and dispatches sub-tasks to concurrent workers on separate desktop instances. On ParaGUIBench, ParaGUI reaches a 46.4% success rate, outperforming the strongest serial baseline (Claude Sonnet 4.6) by 12.9 points while using roughly half the steps and less than half the tokens. These results show that parallel execution can improve both success rate and efficiency on decomposable, long-horizon GUI tasks, pointing to a direction worth further study.

Post-training agents for automated AI research requires optimizing not only model parameters, but also the runtime harness that shapes how research trajectories are generated, evaluated, and learned from. Existing pipelines typically train models under a fixed harness, including prompts, tools, skills, middleware, and memory, while leaving the data-generating process outside the optimization objective. This creates a mismatch between model updates and the static scaffolding that determines trajectory quality. We introduce Co-Harness, a framework that jointly optimizes the agent harness and model parameters during post-training. Co-Harness alternates between harness optimization and model optimization. An LLM-based HarnessCritic analyzes failed trajectories, identifies harness-level failure modes, and proposes validated local updates. The model is then fine-tuned on high-quality trajectories generated by the improved harness, distilling effective scaffolding into model parameters. A 200+ hour autonomous case study further shows that Co-Harness can recover from system crashes, improve inference efficiency, and discover ensemble strategies without human intervention. These results suggest that joint harness and model optimization is an effective way to improve agents beyond fixed-harness post-training.

Monitoring poverty and food security indicators is imperative for addressing socioeconomic challenges in developing nations. A limitation is mismatches in scale between data sources: census data provide geographic coverage, while socioeconomic indicators are derived from infrequently conducted surveys at coarse resolutions, posing a methodological challenge. This study introduces a deep learning framework, JuGAAD, using Indian census and survey data from 2001 and 2011 as a case study. We employ a three-step process: census and geospatial data are averaged into intermediate village-cluster-scale tessellations to reduce noise and regularize administrative boundary changes; an autoencoder compresses high-dimensional National Sample Survey Office (NSSO) data into a low-dimensional latent representation; and a regression model maps upscaled census and geospatial data to this representation. This function is applied to fine-grained census data to generate high-resolution predictions, validated against ground-truth district-level NSSO indicators. Results confirm the methodology predicts socioeconomic indicators at fine scales with strong accuracy.

AI Assistants are increasingly deployed in high-stakes settings, such as healthcare or government services. Yet their real-world behavior remains poorly understood due to strong context dependence. Current evaluation protocols follow a defense-in-depth paradigm with compounding layers of safeguards, ranging from traditional benchmarks to live or adversarial testing. Such benchmarks remain largely static and single-turn, limiting their ability to capture real-world variability in conversational settings. We propose StabilityBench, a principled, general and model-agnostic benchmark operator that transforms single-turn benchmark queries into multi-turn interaction histories. StabilityBench augments existing benchmarks by injecting realistic user simulations, through demographic proxies or sycophantic baits, while preserving original task intent. We apply StabilityBench to four benchmarks spanning mathematical reasoning, health question-answering and safety, and evaluate nine large language models under these conditions. Our results show that model performance is consistently unstable under these injections, with considerable performance degradations on three out of four benchmarks studied. These highlight important limitations of static evaluations and motivate more realistic evaluation settings. To this end, we propose StabilityBench-Mini: a size-preserving variant of StabilityBench that samples across diversification axes, enabling more realistic evaluation without increasing costs.

Scientific discovery is increasingly shifting from isolated disciplines to multi-domain reasoning, and AI for science faces a similar transition. Existing systems are either specialised for individual domains or unify scientific data mainly through text tokenisation and prompt-based interfaces, limiting their ability to handle diverse scientific inputs, produce modality-native outputs, and support joint understanding, reasoning, and generation across scientific domains. We introduce MKB, a unified scientific multimodal model for both understanding and generation, built around a shared Transformer backbone and modality-tailored encoders, adapters, and decoders. MKB covers six scientific branches, including DNA, RNA, proteins, small molecules, earth science, and medical images, and supports native outputs such as biological sequences, molecular strings, meteorological fields, and segmentation masks. Training follows a two-stage modality-then-language curriculum: Stage 1 aligns modality-specific components with the frozen backbone, and Stage 2 consolidates them with the language backbone using mixed scientific and general corpora. Experiments show that MKB achieves competitive scientific understanding across biological and molecular benchmarks, produces high-fidelity native outputs for weather forecasting, biological generation, and medical-image segmentation, and largely retains the general capabilities of its Qwen3-VL backbone. These results demonstrate the feasibility of the proposed paradigm, suggesting that shared-backbone models with modality-tailored components can provide a promising foundation for future cross-domain scientific multimodal exploration. The model and code are publicly available at https://github.com/Shanghai-Academy-of-AI-For-Science/MKB and https://huggingface.co/sais-org/MKB.

In large-scale text analysis tasks, pre-trained language models are often used to embed text corpora for downstream analysis. However, such models may struggle to capture domain-specific semantics and adapting them typically requires large amounts of labeled data and technical expertise to implement training pipelines. Recent approaches have demonstrated how visual interactions in document projections can capture human feedback as training signals for model tuning. However, these methods operate on document-level feedback, which requires users to open and assess individual documents in order to provide effective feedback. In this paper, we propose KeySI, an interaction framework that enables feature-level feedback through keyword-based concept specification. Users specify feedback by organizing extracted keywords into groups representing concepts, which KeySI translates into document-level supervision for subsequent tuning. By operating on keywords as the primary interaction medium, KeySI reduces the need for manual document inspection and labeling and lowers the barrier to adapting embedding models. We present a prototype implementation that, given a corpus, curates representative keywords, visualizes keywords and document embeddings via dimensionality reduction, allows interactive specification of keyword groups, and supports iterative refinement through system feedback. We evaluate KeySI through a user study, usage scenarios, and quantitative experiments demonstrating its effectiveness in capturing user intent and improving embedding alignment.

High-content morphological profiling (Cell Painting) yields sensitive, high-dimensional signatures of cellular state, but translating longitudinal morphology trajectories into interpretable biology remains difficult, especially for weak, chronic perturbations such as low-dose-rate ionizing radiation. Large language models (LLMs) can synthesize heterogeneous evidence into biological narratives, yet their scientific use requires quantitative auditing. We present an evaluation-first, retrieval-augmented interpretation framework for longitudinal Cell Painting morphology, applied to a 9-week RPE-1 time course across five dose rates (0.003–6.0 mGy/hr). Week-matched treated-control morphology deltas are combined with retrieved perturbation neighbors, pathway context, and literature evidence through stable evidence identifiers, enabling an LLM to generate structured, evidence-linked hypotheses that are hierarchically summarized while preserving provenance. We introduce two quantitative auditing tests: V1 citation validity, which verifies that cited evidence identifiers exist in the prompt, and V2 proxy-based morphology compatibility, which evaluates consistency between predicted biological processes and the most altered morphology features. In our experiments, V1 detected no invalid evidence references, while V2 showed meaningful morphology compatibility that increased with perturbation strength and was positively associated with an independent morphology drift summary. The framework produces auditable, falsifiable biological hypotheses, including an adaptive phenotype involving metabolic reprogramming and proteostatic stress at lower dose rates (0.003–0.3 mGy/hr). Current limitations include proxy-based evaluation and the lack of ground-truth mechanism labels.

Tool-augmented language-model agents execute multi-step workflows over external systems, resolving an entity once and then acting on it across subsequent steps. Prior work shows that in single-step actions, agents select the correct tool but bind it to the wrong entity 24-26% of the time. We study what happens to entity bindings over time: do they stay correct, silently drift to a different entity, or, if wrong from the start, propagate and compound? We formalize binding drift (correct at step 1, wrong later) as distinct from error propagation (wrong at step 1, carried forward), and score them on disjoint workflow sets so the two cannot be conflated. In a controlled multi-step testbed (200 workflows, 580 entity-binding-scored steps, four enterprise domains, eight model backends spanning small to frontier), we find: (1) under controlled error injection, an entity lock (the intuitive "persist the first binding" fix) amplifies wrong actions from 907 to 2,746 (3.0x; bootstrap 95% CI [2.8, 3.3]), because it faithfully carries the seeded wrong entity into every later step; (2) the amplification reaches 8.5x on the most affected model (Claude Opus 4.5); (3) a practical LLM-based re-verifier (a single cheap second model call re-reading the original instruction) reduces wrong actions by 79% (0.21x; CI [0.18, 0.25]), closing the gap to within 1 percentage point of an oracle upper-bound (0.20x); and (4) in the natural (non-injected) setting, baseline agents drift on 18% of eligible workflows, with the per-step error rate rising across steps. Persistence and re-verification are not interchangeable: a defense that eliminates drift can worsen propagation, and a practical re-verifier nearly matches oracle recovery.

Comparing phylogenetic tree topologies is essential for understanding epidemic dynamics, yet biologically meaningful distances such as the Subtree Prune and Regraft (SPR) distance are NP-hard to compute and intractable on large datasets. We investigate whether a Graph Neural Network (GNN) can approximate SPR distances in near-constant time per comparison after training. Our contributions are fourfold. First, we build and publicly release a dataset of 864 phylogenetic trees inferred with UPGMA and Neighbor-Joining over four bacterial species, spanning up to 9{,}500 isolates, together with 388 labelled tree pairs. Second, we establish a reproducible pre-processing pipeline including midpoint re-rooting, which reduces tree depth and supplies the rooting required for exact distance computation and for the model's root-based features. Third, we validate the supervision target: on small trees, where exact SPR is tractable, the unrooted phangorn::SPR.dist heuristic correlates almost perfectly with the exact rooted distance computed by rspr (Pearson $0.98$–$0.99$), making it an excellent monotonic surrogate. Lastly, we train a Siamese Graph Isomorphism Network (GIN) regressor. In-distribution, i.e., held-out trees from the same species and size range as training, it explains roughly 87–90% of the variance ($R^2 \approx 0.87$ on a held-out split; $0.90 \pm 0.19$ under stratified cross-validation), with about four times lower error than a mean-predictor baseline, and shows partial transfer to unseen species ($R^2 \approx 0.37$). Its main limitation is extrapolation to trees larger than those seen in training, where accuracy collapses. The released dataset and the validated heuristic versus exact relationship provide a reproducible basis for scaling learned SPR approximation.

Synthetic-population tools increasingly run every individual as an independent large language model (LLM) agent. Using real survey microdata, we show that this paradigm has a basic failure mode, and we set a distribution-first corrective against it, all measured with a deterministic, construct-validated verifier on non-WEIRD (Turkey-first) data. First, N independent LLM agents grounded on 2,414 real World Values Survey respondents fail to reproduce the population's response distribution: they pile onto a modal default (four scenarios x five seeds: concentration 0.36->0.69, entropy 1.46->0.77, 85% collapse, TVD=0.44), and the collapse is a predictable function of scenario structure (r=0.55 with a single-answer structure). Second, Verbalized Sampling (VS) fixes the field's chronic under-dispersion without training in three model families (fidelity +7 to +10; significant on Qwen, p=0.002, d=6.2), yet the same move universally overshoots into over-dispersion (SD-ratio 0.4-0.56 -> 1.26-1.37), a structural property of VS. Third, survey fidelity transfers only weakly to agentic behavior: in a single-model, single-domain booking task, a persona is dominated by a cheapest-default (~80%) that income modulates but does not override (comfort choice 0%->7%->32% across income bands). Fourth, a placebo-controlled memorization attack and an election backtest show VS keeps aggregate strength while subgroup and individual claims are contaminated by recall and underdetermination. We close with the corrective: model the distribution once (VS) and assign it to grounded characters at O(1) cost, with a budget-aware router whose honest AUC is 0.805, not the tautological 1.0 of a code-derived oracle. The central contribution needs no realism claim: it measures the internal inconsistency of the independent-agent route and the conditions under which the distribution-first route calibrates.

Calibration of grey-box simulation models is a constrained optimization problem in which model evaluations are expensive, the parameter space can be high-dimensional, and the search must respect plausibility constraints. Although the simulation code is fully available to the analyst, the joint effect of multiple parameters remains difficult to predict analytically. Classical optimizers such as Nelder–Mead (NM) are simple to deploy but sample-inefficient, particularly under constraints. Modern Bayesian Optimization methods achieve competitive solutions with far fewer evaluations but require non-trivial modeling machinery for constraint handling. We introduce an agentic calibration method in which a large language model acts as the optimizer, with constraints incorporated as a plain-language section of the system prompt. We evaluate the agentic method, NM, and Bayesian Optimization (BO) on an anal cancer simulation model under both unconstrained and clinically constrained calibration. Under unconstrained calibration, the agentic method achieves substantially lower best error than BO and NM, while requiring fewer model evaluations. Under constrained calibration, the agentic method reaches comparable error levels and both outperform NM. These results are obtained at the cost of increased inference time per iteration. Agentic calibration achieves competitive performance with substantially fewer model evaluations, and constraint handling is essentially free at the modeller-facing interface through simple textual specifications rather than additional modelling machinery. The main trade-off lies in increased per-iteration inference cost, making the approach particularly suitable when simulation time dominates. Beyond performance, the per-iteration rationale makes the search auditable and explainable, so its decisions can be scrutinised and justified to third parties.

Sharpness-Aware Minimization (SAM) improves generalization by minimizing the worst-case loss in a local parameter neighborhood. Standard SAM implicitly allocates its global perturbation budget across parameter blocks according to instantaneous minibatch gradient norms. Such an allocation can be noisy and may not reflect the sensitivity that blocks accumulate throughout training. We propose Gradient-Energy Adaptive Radius SAM (GEAR-SAM), which maintains an exponential moving average (EMA) of squared block gradients as a lightweight, curvature-related sensitivity signal and allocates the fixed SAM budget through a closed-form constrained optimization. GEAR-SAM preserves the global SAM radius, requires no Hessian-vector products or explicit Fisher estimation, and adds only scalar state beyond SAM. Experiments on image classification, transfer learning, noisy-label learning, and partition studies demonstrate improved generalization and robustness across architectures and tasks. More broadly, GEAR-SAM provides a dynamic view of sharpness-aware optimization: a fixed perturbation budget should be redistributed as the sensitivity of functional network blocks evolves during training.

Some limits on what language models know are not gaps in data coverage but structural properties of learning from text. We introduce the information shadow: the region of phenomena that a text-trained learner cannot acquire regardless of scale, comprising (I) structures language cannot express, (II) functions that are statistically non-identifiable from the training distribution, and (III) functions that are representable but unreachable by gradient-based training. We give each type a probe that is decisive because the premise of the shadow is, in that setting, provable. For Type I, Language Compression Residuals compare a text learner, which sees only a lossy text-like encoding of the signal, against a full-signal learner, which sees the underlying signal directly. The text learner sits at a computable expressibility ceiling while the full-signal learner pulls away by a gap that stays flat across 300x more data, so the deficit is a property of the channel, not of training. For Type II, the Counterfactual Distinction Test trains models on data exactly consistent with two incompatible rules. Across a provable string task and a language-like agreement task, behavior on counterfactuals is set by the model's inductive bias, while 5% disambiguating data steers the learned rule bidirectionally to either target (r = +/-1.0, p < 1e-10). For Type III, Basin Escape Mapping exhibits a function that is representable at 100% (by hand construction) yet reached 0% of the time by standard training and instantly from a nearby initialization, with width scaling providing no help (p = 1.6 x 10^-14). Each effect is isolated by a control that rules out a capacity or modality artifact. We release the probe suite and discuss implications for benchmark design, capability auditing, and shadow-aware uncertainty.

Mobile devices increasingly rely on real-time NPU inference for camera and perception workloads. Under low-voltage conditions, however, a single inference can induce an instantaneous voltage droop in the power-delivery network, causing the power management integrated circuit to invoke dynamic voltage and frequency scaling (DVFS) and increase latency. We present a measurement study of this effect on a commercial smartphone. We show that aggressive operator fusion in a mobile NPU compiler can create monolithic superlayers whose concentrated execution produces large peak-current bursts. These bursts shift the DVFS-onset voltage upward and reduce the low-voltage operating margin. We further evaluate a practical black-box mitigation: a measurement-guided, pre-compilation graph rewrite that inserts barriers at selected peak-to-average power ratio hot spots to prevent harmful superlayer merging in the vendor NPU compiler. On Snapdragon 8 Gen 3 with MobileNetV4 at 768 x 768 resolution on ImageNet-1k, this method reduces peak current from 3.12 A to 1.94 A with 3.76% latency overhead, preserves stable latency deeper into the low-voltage regime, and shifts the inferred DVFS margin by approximately 173 mV.

In multi-task learning (MTL) negative transfer is often considered as an optimization artifact, but it can also be viewed as a consequence of limited shared capacity and weak task redundancy. We investigate this effect through a Capacity–Redundancy (CR) identity that decomposes the sum of per-task predictive informations into joint predictive information that includes label redundancy defined via total correlation (TC), and a residual coupling term that quantifies interference left unresolved by the shared representation. Additionally, we show two key results: (i) a clustering-gap decomposition that gives a necessary and sufficient condition for clustered sharing to outperform global sharing, and (ii) a gradient–TC bridge in a Gaussian multi-task model that formally justifies gradient cosine similarity as a proxy for redundancy ordering. Empirically, we estimate the residual coupling $\Delta$ from validation residual correlations, showing that clustered LoRA substantially reduces $\widehat{\Delta}$, outperforms size-matched random partitions, and results in statistically significant gains with multi-seed confidence intervals.

Reinforcement learning (RL) has achieved strong results in control, yet learned policies remain brittle to changes in dynamics, action spaces, observation spaces, or goals, a critical limitation for real-world deployment. Existing benchmarks offer limited diversity and complexity, making it difficult to rigorously study transfer, multi-task learning, and meta-learning in RL. We introduce Building2Building (B2B), a large-scale suite of realistic Heating, Ventilation, and Air Conditioning (HVAC) control environments built on EnergyPlus, a state-of-the-art building simulator. B2B is fully compatible with the Gymnasium interface and features a parametric building generator, enabling the systematic generation of diverse building configurations with heterogeneous observation and action spaces. Based on this suite, we define benchmark tasks targeting key open challenges in RL, including goal adaptation, dynamics adaptation, action-space shifts, and cross-domain transfer. By providing a large-scale, diverse, and physically grounded testbed with standardized evaluation protocols, B2B enables systematic investigation of generalization and transfer in continuous control. Beyond advancing research on generalization in RL, this new benchmark also carries significant societal implications by enabling improved HVAC control at scale, one of the most energy-intensive systems in buildings.

As generative AI is increasingly applied to automate multi-step and high-stake workflows, human judgment and involvement remain essential for ensuring the quality of AI-generated outputs. In practice, while it is desirable for human experts to provide oversight on AI regularly, often by reviewing intermediate outputs, giving feedback, making corrections, and steering subsequent steps, such oversight is constrained by the time and resources that humans can afford. This creates a tension between the need for human oversight and AI's efficiency in delivering more output with less intervention. An important but underexplored question, then, is how to optimally engage humans in human-AI coworking. This work was originally motivated by our empirical observation that in long AI workflows, human oversight often improves user satisfaction while reducing unnecessary rework and token consumption. From there, we formulate the problem of where to place oversight stages in human-AI coworking. Under reasonable assumptions, we then develop the nonuniformity principle, which states that the optimal schedule places oversight stages with non-decreasing gaps along the workflow. We empirically validate this principle in two common AI agent workflows: writing literature reviews and constructing websites.

One of the main strengths of Constraint Programming is the ability to reduce the search space via propagation. However, propagation is a double-edged sword, with more pruning power coming at the price of larger computation time. For each problem constraint, the best propagator depends on the specific instance and may change at search time. In the literature, Machine Learning (ML) techniques and activity-based heuristics have been applied respectively for choosing (statically) the propagators for a batch of problems and to adapt (dynamically) the propagation strength. We propose to merge those efforts by using an oracle function, obtained via ML, to decide whether to run complex propagators for a target constraint. A combination of design choices makes the approach flexible and easy to embed in state-of-the-art solvers. In this paper, we focus on investigating the feasibility of building an oracle for the Energetic Reasoning propagator. Our experiments show that high prediction accuracy can be obtained, provide suggestions for classification features, and highlight important issues to address when building such an oracle.