Flash-Moe: Running a 397B Parameter Model on a Mac with 48GB RAM
Flash-MoE enables running a 397 billion parameter Mixture-of-Experts model on a MacBook Pro at 4.4+ tokens/second, utilizing a pure C/Metal inference engine without Python or frameworks.
25 Years of Eggs
AI-driven analysis of 11,345 receipts over 25 years revealed 589 confirmed egg purchases, showcasing the potential of advanced OCR and machine learning techniques in extracting meaningful data from historical documents.
Meta's Omnilingual MT for 1,600 Languages
Omnilingual Machine Translation (OMT) is the first system capable of translating over 1,600 languages, significantly expanding the reach of machine translation beyond the current limitations of 200 languages, thanks to a robust data strategy that includes both public and newly created datasets.
[D] Has industry effectively killed off academic machine learning research in 2026?
Industry dominance in machine learning research has surged, leveraging vast resources and talent, leaving academia to focus on niche topics and theoretical scenarios that lack practical application.
The IBM scientist who rewrote the rules of information just won a Turing Award
Charles H. Bennett, an IBM physicist, has been awarded the 2025 A.M. Turing Award for his groundbreaking work in quantum cryptography, which ensures secure communication based on the laws of physics rather than mathematical complexity.
Sashiko: An agentic Linux kernel code review system
Sashiko is an innovative Linux kernel code review system that automates the evaluation of proposed changes, achieving a 53.6% bug detection rate in tests, significantly outperforming traditional human reviews.
Learnings from training a font recognition model from scratch
Training a font recognition model from scratch revealed that a successful model encompasses more than just a trained file; it requires a comprehensive pipeline for data processing and inference, as demonstrated by the Lens model.
Show HN: A Markdown file that turns your AI agent into an autonomous researcher
Researcher Skill transforms your AI coding agent into a scientific experimenter, autonomously designing and testing over 30 experiments overnight, optimizing code performance with minimal user intervention.
Performance Prediction of Antenna Control Servo System based on LSTM Network [R]
The study explores LSTM networks to enhance the performance of a servo system used in rotating antenna systems for satellite tracking, aiming for improved accuracy and responsiveness.
Tinybox- offline AI device 120B parameters
tinygrad is a rapidly evolving neural network framework that simplifies complex architectures into three core OpTypes: ElementwiseOps, ReduceOps, and MovementOps, enhancing usability for developers.
[D] Solving the "Liquid-Solid Interface" Problem: 116 High-Fidelity Datasets of Coastal Physics (Waves, Saturated Sand, Light Transport)