Search arXiv⌕ Search

arXiv subjects

Artificial intelligence

Search indexed arXiv papers on artificial intelligence and machine learning, including cs.AI metadata. Follow the original manuscripts for methods, experiments and version history.

Follow this topic with RSS
At least 73 records · Page 4Linked to original sources
arXiv2026

Student Guides Teacher: Weak-to-Strong Inference via Spectral Orthogonal Exploration

Large Language Models (LLMs) often suffer from ''Reasoning Collapse'' on challenging mathematical reasoning tasks, where stochastic sampling produces lexical variations of the same erroneous logic rather than genuine semantic exploration. We observe that failed reasoning traces are often associated with a low-rank bias manifold in the model's hidden-state geometry, which reduces exploration toward corrective solution directions. To address this, we propose Spectral Orthogonal Exploration (SOE), a geometric inference framework under a ''Student Guides Teacher'' paradigm. Instead of using a weak auxiliary agent for imitation, SOE uses it as an orthogonal probe to introduce semantically heterogeneous reasoning signals into the teacher's orthogonal complement of its dominant subspace. This intervention steers the teacher toward more diverse reasoning trajectories and improves exploration beyond standard sampling. Experiments on mathematical benchmarks show that SOE improves average accuracy by 62.4\% and average sampling efficiency by 113.7\% over baseline methods, suggesting that geometric interventions can be effective for mitigating reasoning collapse in mathematical reasoning. We further provide preliminary evidence that SOE is also effective on logic and code generation benchmarks.

cs.AI↗
arXiv2026

What Users Leave Unsaid: Under-Specified Queries Limit Vision-Language Models

Current vision-language benchmarks predominantly feature well-structured questions with clear, explicit prompts. However, real user queries are often informal and underspecified. Users naturally leave much unsaid, relying on images to convey context. We introduce HAERAE-Vision, a benchmark of 653 real-world visual questions from Korean online communities (0.76% survival from 86K candidates), each paired with an explicit rewrite, yielding 1,306 query variants in total. Evaluating 39 VLMs, we find that even state-of-the-art models (GPT-5, Gemini 2.5 Pro) achieve under 50% on the original queries. Crucially, query explicitation alone yields 8 to 22 point improvements, with smaller models benefiting most. We further show that even with web search, under-specified queries underperform explicit queries without search, revealing that current retrieval cannot compensate for what users leave unsaid. Our findings demonstrate that a substantial portion of VLM difficulty stem from natural query under-specification instead of model capability, highlighting a critical gap between benchmark evaluation and real-world deployment.

cs.CV↗
arXiv2026

The Psychology of Learning from Machines: Anthropomorphic AI and the Paradox of Automation in Education

As AI tutors enter classrooms at unprecedented speed, their deployment increasingly outpaces our grasp of the psychological and social consequences of such technology. Yet decades of research in automation psychology, human factors, and human-computer interaction provide crucial insights that remain underutilized in educational AI design. This work synthesizes four research traditions -- automation psychology, human factors engineering, HCI, and philosophy of technology -- to establish a comprehensive framework for understanding how learners psychologically relate to anthropomorphic AI tutors. We identify three persistent challenges intensified by Generative AI's conversational fluency. First, learners exhibit dual trust calibration failures -- automation bias (uncritical acceptance) and algorithm aversion (excessive rejection after errors) -- with an expertise paradox where novices overrely while experts underrely. Second, while anthropomorphic design enhances engagement, it can distract from learning and foster harmful emotional attachment. Third, automation ironies persist: systems meant to aid cognition introduce designer errors, degrade skills through disuse, and create monitoring burdens humans perform poorly. We ground this theoretical synthesis through comparative analysis of over 104,984 YouTube comments across AI-generated philosophical debates and human-created engineering tutorials, revealing domain-dependent trust patterns and strong anthropomorphic projection despite minimal cues. For engineering education, our synthesis mandates differentiated approaches: AI tutoring for technical foundations where automation bias is manageable through proper scaffolding, but human facilitation for design, ethics, and professional judgment where tacit knowledge transmission proves irreplaceable.

cs.CY↗
arXiv2026

Political Alignment in Large Language Models: A Multidimensional Audit of Psychometric Identity and Behavioral Bias

As large language models (LLMs) are increasingly deployed, understanding how they express political positioning is important for evaluating alignment and downstream effects. We audit 26 contemporary LLMs using three political psychometric inventories (Political Compass, SapplyValues, 8Values) and a news bias labeling task. To test robustness, inventories are administered across multiple semantic prompt variants and analyzed with a two-way ANOVA separating model and prompt effects. Most models cluster in a similar ideological region, with 96.3% located in the Libertarian-Left quadrant of the Political Compass, and model identity explaining most variance across prompt variants ($η^2 > 0.90$). Cross-instrument comparisons suggest that the Political Compass social axis aligns more strongly with cultural progressivism than authority-related measures ($r=-0.64$). We observe differences between open-weight and closed-source models and asymmetric performance in detecting extreme political bias in downstream classification. Regression analysis finds that psychometric ideological positioning does not significantly predict classification errors, providing no evidence of a statistically significant relationship between conversational ideological identity and task-level behavior. These findings suggest that single-axis evaluations are insufficient and that multidimensional auditing frameworks are important to characterize alignment behavior in deployed LLMs. Our code and data are publicly available at https://github.com/sakhadib/PolAlignLLM.

cs.CY↗
arXiv2026

Geometry-Aware Hallucination Detection in Large Language Models

Large language models (LLMs) frequently generate factually incorrect or unsupported content, commonly referred to as hallucinations. Prior work has explored decoding strategies, retrieval augmentation, and supervised fine-tuning for hallucination detection, while recent studies show that in-context learning (ICL) can substantially influence factual reliability. However, existing ICL demonstration selection methods often rely on surface-level similarity heuristics and exhibit limited robustness across tasks and models. We propose GA-ICL, a geometry-aware demonstration sampling framework for selecting in-context demonstrations that leverages latent representations extracted from frozen LLMs. By jointly modeling local manifold structure and class-aware prototype geometry, GA-ICL selects demonstrations based on their proximity to learned prototypes rather than lexical or embedding similarity alone. Across factual verification (FEVER) and hallucination detection (HaluEval) benchmarks, GA-ICL outperforms standard ICL selection baselines in the majority of evaluated settings, with particularly strong gains on dialogue and summarization tasks. The method remains robust under temperature perturbations and model variation, indicating improved stability compared to heuristic retrieval strategies. While lexical retrieval can remain competitive in certain question-answering regimes at smaller model scales, our results demonstrate that geometry-aware prototype selection provides a reliable and training-light approach for hallucination detection without modifying LLM parameters. Extended evaluations on Phi-14B and Qwen3-32B confirm that GA-ICL scales effectively to larger models, outperforming all compared baselines including on QA tasks where smaller models show boundary-condition limitations, offering a principled direction for improved ICL demonstration selection.

cs.LG↗
arXiv2026

Akasha 2: Hamiltonian State Space Duality and Visual-Language Joint Embedding Predictive Architectur

We present Akasha 2, a state-of-the-art multimodal architecture that integrates Hamiltonian State Space Duality (H-SSD) with Visual-Language Joint Embedding Predictive Architecture (VL-JEPA). The system leverages the Mamba-3 Selective State Space Model (SSM) augmented by a Sparse Mixture of Hamiltonian Experts (SMoE-HE) that enforces latent physical conservation laws through symplectic integration. For visual synthesis, we introduce Hamiltonian Flow Matching (HFM) and persistent 3D Gaussian Splatting (3DGS), enabling ultra-low latency (<50ms) on mobile hardware. This work establishes a new paradigm in latent world models, achieving unprecedented spatiotemporal coherence through a holographic memory architecture. Our approach demonstrates that incorporating physics-inspired inductive biases into neural architectures yields significant improvements: state-of-the-art video prediction (FVD: 287), 4x faster visual synthesis than diffusion models, and 3-18x inference speedup over transformer baselines while maintaining energy conservation over extended horizons.

cs.CV↗
arXiv2026

When Smaller Wins: Dual-Stage Distillation and Pareto-Guided Compression of Liquid Neural Networks for Edge Battery Prognostics

Battery management systems increasingly require accurate battery health prognostics under strict on-device constraints. This paper presents DLNet, a practical framework with dual-stage distillation of liquid neural networks that turns a high-capacity model into compact and edge-deployable models for battery health prediction. DLNet first applies Euler discretization to reformulate liquid dynamics for embedded compatibility. It then performs dual-stage knowledge distillation to transfer the teacher model's temporal behavior and recover it after further compression. Pareto-guided selection under joint error-cost objectives retains student models that balance accuracy and efficiency. We evaluate DLNet on a widely used dataset and validate real-device feasibility on an Arduino Nano 33 BLE Sense using int8 deployment. The final deployed student achieves a low error of 0.0066 when predicting battery health over the next 100 cycles, which is 15.4% lower than the teacher model. It reduces the model size from 616 kB to 94 kB with 84.7% reduction and takes 21 ms per inference on the device. These results support a practical smaller wins observation that a small model can match or exceed a large teacher for edge-based prognostics with proper supervision and selection. Beyond batteries, the DLNet framework can extend to other industrial analytics tasks with strict hardware constraints.

cs.LG↗
arXiv2026

Automated QoR improvement in OpenROAD with coding agents

EDA development and innovation has been constrained by scarcity of expert engineering resources. While leading LLMs have demonstrated excellent performance in coding and scientific reasoning tasks, their capacity to advance EDA technology itself has been largely untested. We present AuDoPEDA, an autonomous, repository-grounded coding system built atop OpenAI models and a Codex-class agent that reads OpenROAD, proposes research directions, expands them into implementation steps, and submits executable diffs. Our contributions include (i) a closed-loop LLM framework for EDA code changes; (ii) a task suite and evaluation protocol on OpenROAD for PPA-oriented improvements; and (iii) end-to-end demonstrations with minimal human oversight. Experiments in OpenROAD achieve routed wirelength reductions of up to 5.9%, effective clock period reductions of up to 10.0%, and power reductions of up to 19.4%.

cs.SE↗
arXiv2026

$\texttt{AMEND++}$: Benchmarking Eligibility Criteria Amendments in Clinical Trials

Clinical trial amendments frequently introduce delays, increased costs, and administrative burden, with eligibility criteria being the most commonly amended component. We introduce \textit{eligibility criteria amendment prediction}, a novel NLP task that aims to forecast whether the eligibility criteria of an initial trial protocol will undergo future amendments. To support this task, we release $\texttt{AMEND++}$, a benchmark suite comprising two datasets: $\texttt{AMEND}$, which captures eligibility-criteria version histories and amendment labels from public clinical trials, and $\verb|AMEND_LLM|$, a refined subset curated using an LLM-based denoising pipeline to isolate substantive changes. We further propose $\textit{Change-Aware Masked Language Modeling}$ (CAMLM), a revision-aware pretraining strategy that leverages historical edits to learn amendment-sensitive representations. Experiments across diverse baselines show that CAMLM consistently improves amendment prediction, enabling more robust and cost-effective clinical trial design.

cs.CL↗
arXiv2026

C-World: A Computer Use Agent Environment Creator

To close the gap between LLM-based agents and humans in planning and reasoning, agents need large-scale, diverse environments for continuous learning -- yet building such environments is itself prohibitively expensive. We present C-World, an environment creation system that enables users to build agent environments on demand. We define a complete agent environment through four components: an Action Space of 5,571 format-unified tools across 204 common applications, a Task Distribution engine that synthesizes long-horizon workflows with wild constraints, a Transition Function implemented as a state controller that injects realistic failures and perturbations, and a Reward Signal combining verifiable metrics with LLM-based judgment. C-World operates in two modes: a realistic mode grounded in live API execution, and a synthesized mode powered by the World Engine, which approximates tool behavior without live service access, enabling scalable environment creation -- including environments for domains and tools that do not yet exist in the real world. Evaluation of nine state-of-the-art LLMs reveals that planning ability is uniformly strong but execution remains the bottleneck, and that constraint following -- not tool invocation -- is the dominant failure mode. The World Engine achieves Spearman $ρ= 0.883$ ranking correlation with real execution, and fine-tuning on just 1,170 C-World trajectories outperforms baselines trained on 119k samples, demonstrating C-World's dual value as a rigorous evaluation environment and a scalable data engine. Our code and data are available at https://ziqiao-git.github.io/C-World/

cs.AI↗
arXiv2026

On the Fallacy of Global Token Perplexity in Spoken Language Model Evaluation

Generative spoken language models pretrained on large-scale raw audio can continue a speech prompt with appropriate content while preserving attributes like speaker and emotion, serving as foundation models for spoken dialogue. In prior literature, these models are often evaluated using ``global token perplexity'', which directly applies the text perplexity formulation to speech tokens. However, this practice overlooks fundamental differences between speech and text modalities, possibly leading to an underestimation of the speech characteristics. In this work, we propose a variety of likelihood- and generative-based evaluation methods that serve in place of naive global token perplexity. We demonstrate that the proposed evaluations more faithfully reflect perceived generation quality, as evidenced by stronger correlations with human-rated mean opinion scores (MOS). When assessed under the new metrics, the relative performance landscape of spoken language models is reshaped, revealing a significantly reduced gap between the best-performing model and the human topline. Together, these results suggest that appropriate evaluation is critical for accurately assessing progress in spoken language modeling.

cs.CL↗
arXiv2026

Future-as-Label: Scalable Supervision from Real-World Outcomes

Time creates free supervision: forecasts about real-world events resolve to verifiable outcomes. The passage of time provides labels that require no annotation. To exploit this structure, we extend reinforcement learning with verifiable rewards to real-world prediction over time. We train language models to make probabilistic forecasts from causally masked information, using proper scoring rules as the reward function once events resolve. Learning is driven entirely by realized outcomes, enabling scalable outcome-based supervision in open-world prediction. On real-world forecasting benchmarks, Qwen3-32B trained using Foresight Learning improves Brier score by 27% and halves calibration error relative to its pretrained baseline, and outperforms Qwen3-235B on both constructed future-event prediction tasks and the Metaculus benchmark despite a 7x parameter disadvantage.

cs.LG↗
arXiv2026

Circuit Mechanisms for Spatial Relation Generation in Diffusion Transformers

Diffusion Transformers (DiTs) have greatly advanced text-to-image generation, but models still struggle to generate the correct spatial relations between objects as specified in the text prompt. In this study, we adopt a mechanistic interpretability approach to investigate how a DiT can generate correct spatial relations between objects. We train, from scratch, DiTs of different sizes with different text encoders to learn to generate images containing two objects whose attributes and spatial relations are specified in the text prompt. We find that, although all the models can learn this task to near-perfect accuracy, the underlying mechanisms differ drastically depending on the choice of text encoder. When using random text embeddings, we find that the spatial-relation information is passed to image tokens through a two-stage circuit, involving two cross-attention heads that separately read the spatial relation and single-object attributes in the text prompt. When using a pretrained text encoder (T5), we find that the DiT uses a different circuit that leverages information fusion in the text tokens, reading spatial-relation and single-object information together from a single text token. We further show that, although the in-domain performance is similar for the two settings, their robustness to out-of-domain perturbations differs, potentially suggesting the difficulty of generating correct relations in real-world scenarios.

cs.AI↗
arXiv2026

CARD: Cluster-level Adaptation with Reward-guided Decoding for Personalized Text Generation

Adapting large language models to individual users remains challenging due to the tension between fine-grained personalization and scalable deployment. We present CARD, a hierarchical framework that achieves effective personalization through progressive refinement. CARD first clusters users according to shared stylistic patterns and learns group-specific LoRA adapters, enabling robust generalization and strong low-resource performance. To capture individual differences within each cluster, we propose an implicit preference learning mechanism that contrasts user-authored text with cluster-level generations, allowing the model to infer user-specific style preferences without manual annotation. At inference time, CARD injects personalization exclusively at decoding via lightweight user preference vectors and low-rank logit corrections, while keeping the base model frozen. Experiments on the LaMP and LongLaMP benchmarks show that CARD achieves superior generation quality compared to baselines, while significantly improving efficiency and scalability for practical personalized text generation.

cs.AI↗
arXiv2026

Do Implicit Personalization and Explicit Styles Conflict? PsPLUG: A Lightweight Plug-in for Balancing Personalization and Style in Customized LLMs

Personalized large language models are often expected to follow explicit style instructions, yet we find that such instructions can undermine the user-specific characteristics that personalization methods aim to preserve. We call this failure mode personalization collapse: explicit style control can conflict with implicit user preferences. To address this challenge, we propose PsPLUG, a lightweight plug-in that learns a user-specific residual after accounting for the requested style. PsPLUG also allows us to tune personalization strength at inference time. Our experiments show that explicit style instructions can diminish personalization in existing methods, whereas PsPLUG better preserves user preferences while providing precise control over the balance between personalization and style adherence.

cs.AI↗
arXiv2026

SafeGPT: Preventing Data Leakage and Unethical Outputs in Enterprise LLM Use

Large Language Models (LLMs) are transforming enterprise workflows but introduce security and ethics challenges when employees inadvertently share confidential data or generate policy-violating content. This paper proposes SafeGPT, a two-sided guardrail system preventing sensitive data leakage and unethical outputs. SafeGPT integrates input-side detection/redaction, output-side moderation/reframing, and human-in-the-loop feedback. Experiments demonstrate SafeGPT effectively reduces data leakage risk and biased outputs while maintaining satisfaction.

cs.CR↗
arXiv2026

Context Matters: Peer-Aware Student Behavioral Engagement Measurement via VLM Action Parsing and LLM Sequence Classification

Understanding student behavior in the classroom is essential to improve both pedagogical quality and student engagement. Existing methods for predicting student engagement typically require substantial annotated data to model the diversity of student behaviors, yet privacy concerns often restrict researchers to their own proprietary datasets. Moreover, the classroom context, represented in peers' actions, is ignored. To address the aforementioned limitation, we propose a novel three-stage framework for video-based student engagement measurement. First, we explore the few-shot adaptation of the vision-language model for student action recognition, which is fine-tuned to distinguish among action categories with a few training samples. Second, to handle continuous and unpredictable student actions, we utilize the sliding temporal window technique to divide each student's 2-minute-long video into non-overlapping segments. Each segment is assigned an action category via the fine-tuned VLM model, generating a sequence of action predictions. Finally, we leverage the large language model to classify this entire sequence of actions, together with the classroom context, as belonging to an engaged or disengaged student. The experimental results demonstrate the effectiveness of the proposed approach in identifying student engagement. The source code will be available at https://github.com/ahmed-nady/context_aware_student_engagement.

cs.CV↗
arXiv2026

BizFinBench.v2: Towards Reliable LLMs in Finance via Real-User Data and Offline/Online Bilingual Evaluation

Large language models are becoming increasingly significant in financial applications. Nevertheless, prevailing benchmarks are largely dependent on simulated or generic data, which leads to a significant gap between reported performance and actual efficacy in real-world scenarios. To tackle this challenge, we present BizFinBench.v2, the first integrated offline and online benchmark built upon authentic user query-response data from both Chinese and U.S. equity markets. It comprises 28,860 questions across eight offline and two online tasks. Experimental results show that GPT-5 achieves a mere 61.5% accuracy, still failing to meet the practical business requirement (84.8%). Among the evaluated commercial models, DeepSeek-R1 exhibits superior investment efficacy. Error analysis grounded in real financial practice reveals persistent limitations in existing models. By overcoming the constraints of prior benchmarks, BizFinBench.v2 provides a substantiated foundation for advancing LLM deployment in the financial sector. Our data and code are available at https://github.com/HiThink-Research/BizFinBench.v2.

cs.AI↗