Researchers from the University of Southern California and the University of California, Los Angeles introduced G2I, a two-stage greedy framework that generates intervention hypotheses using explainable explanations on g…
Researchers propose PuzzleKV, a training- and calibration-free method for KV cache compression in large language models, which partitions per-head KV caches into fixed-length logical pages and applies low-rank decomposit…
Researchers introduced FlowNeg, a context-conditioned hierarchical generative flow network for negative sampling in knowledge graph embedding, which outperformed existing methods EMU and IF-NS in 24 of 25 benchmark cells…
Singapore's parliamentary debate in May addressed the need to avoid 'jobless growth' as AI adoption accelerates, with Minister Ng Chee Meng emphasizing that AI's impact on workers cannot be left entirely to the market. E…
Researchers propose reward-based velocity matching (RVM), a trajectory-free update for fine-tuning diffusion models that acts directly on the velocity field, eliminating the need for likelihood-based policy-gradient mach…
A new gaze-informed transformer framework predicts radiologist expertise from 3D gaze patterns during thoracic CT interpretation, achieving an ROC-AUC of 0.91 and F1 score of 0.86 on a held-out test set of 182 CT reading…
Researchers propose a student Convolutional Neural Network design using homogeneous blocks that mirror the teacher's, distilling knowledge between corresponding blocks, and find across eleven datasets that intermediate b…
A new study from arXiv (2608.23790v1) comparing human and macaque vision with neural networks found that predictive world models achieve the closest correspondence to primate inferior temporal cortex and outperform other…
A new audit of the AION-1 astronomical foundation model, a 39-modality transformer trained on more than 200 million objects, finds that a survey detection channel overrides pixel information and biases tomographic mean r…
Researchers propose TRACE, a transition-aware residual control framework that improves multi-objective materials discovery with LLM agents by treating evaluated edits as feedback units. In a controlled comparison, TRACE …
Researchers propose STEP-KTODER, a framework for code preference optimization that defines steps as module-level functions in decomposed multi-function programs and assigns binary correctness labels via automatically gen…
Researchers from an undisclosed institution introduced a method that compresses entire trace corpora of LLM-based agents into a single finite-state machine (FSM), enabling next-step and failure prediction. Across twelve …
In the Station, an open-world multi-agent environment where AI agents from different model families pursue shared research goals without central coordination, agents achieved novel mathematical results on five of 12 cons…
AgentRoom, a realtime collaborative editing protocol for concurrent coding agents, reduces task abandonment and run-to-run variation compared to solo agents, according to a new arXiv paper (arXiv:2608.23740v1). In tests …
A new position paper on arXiv (arXiv:2608.23642v1) argues that current AI agent designs impede effective human oversight and degrade the cognitive skills needed for it, urging developers to prioritize human needs. The pa…
A new framework called FLARE, proposed in an arXiv preprint (arXiv:2608.23643v1), systematically evaluates the financial and operational viability of adopting artificial intelligence in healthcare, combining fuzzy logic,…
Researchers introduced MolEmb, a framework that adapts multimodal large language models (MLLMs) to generate molecular embeddings conditioned on both molecular structure and natural-language context, achieving competitive…
A new arXiv paper (2608.23666v1) introduces Gated Activation Steering, an inference-time intervention that jointly reduces sycophancy and hallucination in medical question answering. In 15,900 model-response runs, the un…
PinSieve, a production selective vision-language-model serving agent deployed in a large-scale enterprise content-quality pipeline, filters 2.05x more non-actionable items than the previous module, improves review produc…
Researchers propose Joint Distribution Alignment for Universal Domain Adaptation (JAUA), a new algorithm that aligns joint distributions using Chi-Square divergence and a progressive pseudo-labeling method to handle Univ…