AI/ML arXiv cs.AI

Under What Conditions Can a Machine Be Called Genuinely Creative?

A theoretical paper defining the requirements for genuine machine creativity using a framework from Designics.

AI/ML arXiv cs.AI

MiniMax Sparse Attention

Introduction of MiniMax Sparse Attention (MSA), a blockwise sparse attention mechanism that significantly reduces compute and increases speed for long-context LLMs.

AI/ML arXiv cs.AI

Is It You or Your Environment? A Bayesian Inference Framework for Genomically-Anchored Personalized Physiological Interpretation

A Bayesian inference framework that uses genomic profiles as priors to provide personalized physiological interpretation for health AI.

AI/ML arXiv cs.AI

EurekAgent: Agent Environment Engineering is All You Need For Autonomous Scientific Discovery

Introduction of EurekAgent, an environment-engineered agent system that optimizes autonomous scientific discovery through constrained execution and artifact management.

AI/ML arXiv cs.AI

Application of Artificial Intelligence and Machine Learning in Libraries: A Systematic Review

A systematic review of the application of AI and ML within the library science domain.

AI/ML arXiv cs.AI

Learning Developmental Scaffoldings to Guide Self-Organisation

Researchers propose a model that jointly learns self-organisation rules and pre-patterns using Neural Cellular Automata (NCA) and SIREN to improve robustness and encoding capacity in biological development simulation.

AI/ML arXiv cs.AI

Planning with the Views via Scene Self-Exploration

The ViewSuite framework introduces iterative self-exploration and view graph distillation to help VLMs plan complex multi-turn camera movements in 3D environments.

AI/ML arXiv cs.AI

VikingMem: A Memory Base Management System for Stateful LLM-based Applications

VikingMem is introduced as a memory base management system built on VikingDB to provide stateful, long-term interaction memory for LLM-based applications through selective extraction and temporal compression.

AI/ML arXiv cs.AI

Evidence-Gated LLM Priors for Multi-Objective Bayesian Optimization

This paper proposes an objective-wise reputation-market mechanism to calibrate LLM-generated expert priors in multi-objective Bayesian optimization, reducing reliance on uncalibrated LLM confidence.

AI/ML arXiv cs.AI

EvoTrainer: Co-Evolving LLM Policies and Training Harnesses for Autonomous Agentic Reinforcement Learning

EvoTrainer is an autonomous training framework that co-evolves LLM policies and training harnesses through empirical feedback to improve performance in agentic RL, particularly for software engineering tasks.

AI/ML arXiv cs.AI

Output Type Before Quality: A Standards-Derived XAI Admissibility Rubric for Autonomous-Driving Safety

The authors derive an XAI admissibility rubric for autonomous driving safety, arguing that Causal XAI is structurally required for hazard identification and incident investigation to meet safety standards.

AI/ML arXiv cs.AI

StainFlow: Entity-Stain Tracking and Evidence Linking for Process Rewards in GUI Agents

StainFlow introduces an entity-stain-flow process reward model for GUI agents to provide finer-grained credit assignment in RL by tracking entity states and evidence linking.

AI/ML arXiv cs.AI

Think Fast: Estimating No-CoT Task-Completion Time Horizons of Frontier AI Models

Researchers estimate the task-completion time horizons of frontier AI models without Chain-of-Thought (CoT), finding that internal reasoning capabilities are doubling roughly every year.

Other arXiv cs.AI

Large-scale semantic mapping of learner agency and autonomy reveals what measurement and generative AI research overlook

A semantic analysis of learner agency and autonomy in education reveals that current generative AI research focuses too heavily on learning regulation while overlooking sociocultural dimensions.

AI/ML arXiv cs.AI

GUITrans2Act: Understanding User Operational Behaviors from Mobile GUI Interactions with Vision-Language Models

The Teach VLM model and Teach-and-Repeat paradigm translate mobile screen trajectories into operational knowledge to guide downstream GUI execution agents.

Software Engineering Hacker News

The ghost domain problem in DNS, and what we're doing about it

A discussion on the 'ghost domain' problem in DNS, focusing on how orphaned records can cause resolution issues and the efforts to mitigate them.

AI/ML arXiv cs.AI

Verbatim Chunks Beat Extracted Artifacts: A Controlled Ablation of Memory Representations for Long LLM Conversations

Research showing that keeping verbatim conversation chunks in LLM memory outperforms structured artifacts, as distillation often loses critical detail.

AI/ML arXiv cs.AI

Token-Level LLM Collaboration via FusionRoute

Introduces FusionRoute, a token-level multi-LLM collaboration framework that uses a lightweight router and complementary logits to optimize decoding.

AI/ML arXiv cs.AI

Actionable Interpretability Must Be Defined in Terms of Symmetries

A theoretical paper arguing that AI interpretability must be defined through symmetries to be formally testable and actionable.

AI/ML arXiv cs.AI

Optimizing Agentic Reasoning with Retrieval via Synthetic Semantic Information Gain Reward

Presents InfoReasoner, a framework that uses synthetic semantic information gain rewards to optimize retrieval in agentic reasoning models.