All Articles
16860 articles total
REST: Receding Horizon Explorative Steiner Tree for Zero-Shot Object-Goal Navigation
REST (Receding Horizon Explorative Steiner Tree) is a training-free framework for zero-shot object-goal navigation using a path-tree option space and LLM reasoning.
Show HN: Tarit – a hypervisor which is 2x faster than firecracker
Tarit is a new hypervisor claiming to be twice as fast as Firecracker, targeting high-performance virtualization.
US seeks share of Korean chipmakers' 'excess profits'
The US government is seeking a share of 'excess profits' from South Korean chipmakers, indicating geopolitical tension in the semiconductor industry.
Databricks hits $188B valuation, extending its run as AI’s favorite second act
Databricks has reached a valuation of $188B, pivoting heavily into AI and publishing research on the cost-efficiency of open weight models.
Taylor Farms pulls iceberg lettuce from the US market after cyclosporiasis outbreak
Taylor Farms is recalling iceberg lettuce from the US market following a cyclosporiasis outbreak.
Language as a Wave Phenomenon: Semantic Phase Locking and Interference in Neural Networks
Researchers introduce PRISM, a complex-valued architecture that uses phase-based spectral interference to improve neural sequence modeling efficiency.
Energy-Efficient Federated Learning via Adaptive Encoder Freezing for MRI-to-CT Conversion: A Green AI-Guided Research
A new adaptive encoder freezing strategy for Federated Learning is proposed to reduce energy consumption in MRI-to-CT conversion tasks.
JEEVHITAA -- An HCAI Ecosystem to Support Collective Care
JEEVHITAA is an HCAI ecosystem for collective health care that uses role-aware sharing and RAG-based LLM summaries.
CluCERT: Certifying LLM Robustness via Clustering-Guided Denoising Smoothing
CluCERT is introduced as a framework for certifying LLM robustness against adversarial prompts using clustering-guided denoising smoothing.
Step-Tagging: Toward controlling the generation of Language Reasoning Models through step monitoring
The Step-Tagging framework introduces real-time annotation of reasoning steps in LRMs to enable efficient early stopping and token reduction.
Native Extrapolation Awareness in Flow-Based Conditional Generation
Diverging Flows is a new approach to Flow Matching that detects extrapolation hazards in safety-critical generative models.
Thanks HN for 15 years of support and helping me find my life's work
A user expresses gratitude to the Hacker News community for supporting them over the last 15 years and helping them find their life's work.
Everybody's Weirded Out by AI–Except the People Who Foist It on Us
An opinion piece discussing the societal discomfort with AI and the people who push its adoption.
FAA lets Boeing sign off on 737 MAX, 787 airworthiness certificates again
The FAA has permitted Boeing to resume signing off on the airworthiness certificates for the 737 MAX and 787 aircraft.
The Zoom hack that says, ‘Don’t record me’
An exploration of the social and productivity implications of the ubiquity of AI transcription and summarization in meetings.
SEMA: a Scalable and Efficient Mamba like Attention via Token Localization and Averaging
Introduction of SEMA, a scalable Mamba-like attention mechanism designed to improve efficiency and focus in computer vision tasks.
Similarity as Reward Alignment: Robust and Versatile Preference-based Reinforcement Learning
Presentation of SARA, a contrastive framework for preference-based reinforcement learning that is robust to labeler errors.
AgenticData: An Agentic Data Analytics System for Heterogeneous Data
Introduction of AgenticData, a multi-agent system that allows users to analyze heterogeneous data using natural language queries.
EEG-based AI-BCI Wheelchair Advancement: Transformer-Based Learning with Motor Imagery for Brain Computer Interface
Development of TFormerEEG, a Transformer-based architecture for EEG classification to control AI-BCI wheelchairs.
Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity
Proposal of Verbalized Sampling, a training-free prompting strategy to mitigate mode collapse and increase diversity in LLM responses.