Safety & Evaluation A

The paper proposes a quantum ring all-reduce scheme for distributed learning, arguing that quantum communication can make distributed training both more communication-efficient and information-theoretically private. The approach is discussed for both classical and quantum settings.

The paper presents a model-driven approach for developing families of reinforcement learning environments. It treats virtual training environments as software-intensive systems and aims to make building these safe, cost-efficient alternatives to real-world training more systematic.

The article investigates the key features and surprises of deep learning's training and generalization from a physics-informed perspective. It examines how deep learning departs from classical statistical intuitions to achieve strong real-world performance, justifying these observations where possible.

The paper proposes a shifting-based method for constructing optimizable linear relaxations of general activation functions, used in formal verification of neural networks. It removes the need for hand-crafted relaxations tailored to each activation, supporting formal guarantees in safety- and security-critical settings.

The paper presents PsyScore, a psychometrically-aware framework for automated essay scoring that adapts to writing traits and provides ZPD-scaffolded feedback. It aims to unify scoring and feedback, which existing methods treat separately, balancing reliable assessment with interpretable, actionable instruction.

LLM-driven information services are reshaping how public knowledge is produced. This work proposes a participatory approach to engage editorial expertise in LLM-mediated knowledge dissemination.

This work introduces the Meaning Intelligence Framework, a nine-dimension annotation and evaluation scheme, to study the register gap in Nigerian public discourse.

Vision-Language-Action models pre-trained on massive video-robot datasets have transformed robot control. This work shows that finetuning them requires fewer layers than previously assumed.

Academic paper search is a core step in research, and LLM-based search agents are emerging. ScholarQuest provides a taxonomy-guided benchmark for agentic academic paper search in open literature environments.

Large language models have advanced in reasoning, especially deduction. QMFOL benchmarks LLM reasoning through quantifiable monadic first-order logic test-case generation.

The paper investigates whether emergent misalignment—induced by fine-tuning language models on insecure code—corresponds to a causally actionable, shared direction in activation space. Across four instruction-tuned model families, it studies using such directions to detect and mitigate the misalignment.

Machine learning deployed in evolving streaming environments must handle non-stationarity. This work analyzes and evaluates learner-based approaches to concept drift detection.

The paper presents CzechDocs, a multiway parallel dataset of formatted documents in HTML, DOCX, and PDF covering Czech and minority languages used in Czechia—primarily Ukrainian and English, with smaller amounts of Vietnamese, Russian, and others. It is designed to support evaluation of machine translation.

Machine learning models are mostly evaluated through predictive metrics such as accuracy. This work goes beyond accuracy to measure the logical compliance of predictive models.

The paper studies off-policy evaluation in finite-horizon MDPs when rewards are missing not at random, as in offline reinforcement learning with sparse, irregular, or censored reward records. It develops missingness-aware policies for settings such as health care and marketing.

Psychological instruments designed for humans are increasingly applied to large language models. This work argues that the apparent psychological profiles of LLMs are largely a measurement artifact.

The paper presents an algorithm for pitch spelling and key estimation across jazz lead sheets, solo transcriptions, classical piano, and monophonic scores. Given MIDI-like input with note pitches and bar boundaries, it estimates note names, a global key signature, and local scales.

Using vision foundation models for camera-to-LiDAR knowledge distillation is promising. HilDA advances self-supervised LiDAR pre-training through hierarchical distillation with diffusion.

The paper presents ReNikud, an audio-supervised approach to grapheme-to-phoneme conversion for Modern Hebrew. It addresses the ambiguity of Hebrew's abjad script, which leaves vowels largely unwritten, going beyond standard pipelines that first predict vowel diacritics (nikud).

The paper introduces NAMESAKES, a study probing identity memorization in text-to-image models, which can generate realistic likenesses of individuals from their names. It addresses the difficulty of telling whether a generated face is memorized or fabricated without ground-truth photos, training data, or white-box model access.

The paper develops an adaptive LLM-based high-school tutoring system with subject-aware prompting, built on 14 pedagogical features—such as tutor scaffolding and student understanding—extracted from transcripts. It aims to improve student engagement where static-prompt tutoring struggles to adapt across disciplines.

The paper characterizes when streaming tool use helps in streaming retrieval-augmented generation, which issues tool queries in parallel with ongoing user input to cut perceived latency. It argues the benefit is query-intrinsic and studies how tool intent stabilizes before an utterance is complete.

The paper tests whether large language models resist valid corrections to their own writing during verifiable instruction-following revision. Across four models under genuine authorship, it finds that the documented self-preference bias is weak or absent in this revision setting.

The paper presents IHUBERT, a monolingual Persian pretrained language model trained from scratch on a RoBERTa-base encoder. It uses vector-based semantic deduplication and domain-balanced pretraining to address the scarcity of large, high-quality Persian corpora and limited evaluation.

OpenAI says GPT-5.5 Instant strengthens ChatGPT's health and wellness responses through better reasoning, richer context, and clearer communication. The work is backed by physician-informed evaluations aimed at delivering more reliable, trustworthy health guidance.

The paper proposes source-grounded data generation for text-to-JSON learning, where models extract information from long unstructured documents into structured, machine-readable JSON. It targets domains such as financial filings and clinical records that store high-value information in unstructured text.

The paper investigates over-privileged tool selection in LLM agents, which autonomously choose among tools with different privilege levels. It addresses a gap in prior tool-selection research, which focuses on safety-agnostic metadata preferences, by studying when lower-privilege tools would suffice.

The paper presents Connect the Dots (CoD), a reinforcement-learning framework for training large language models as long-lifecycle agents. It targets the meta-capability of solving a long sequence of tasks while continuously exploring an environment, aiming for cross-domain generalization.

The paper investigates discrepancies between human judgments and MOS prediction models in speech quality assessment, using controlled acoustic and prosodic perturbations. It probes whether these models, widely used as proxy metrics in text-to-speech research, capture quality differences beyond acoustic fidelity.

The paper proposes a lightweight framework for automated pronunciation assessment based on discrete speech token surprisal, trained only on native speech resources. It operates unsupervised or with light calibration from a small set of scored utterances, avoiding costly labeled learner-error or non-native corpora.