AI Briefing
Hacker News news
The latest AI and developer news from Hacker News, with short summaries.
Visit website →Godot will no longer accept AI-written code contributions
The open-source game engine Godot announced a policy banning PR submissions of AI-generated code and agents.
Matrix Orthogonalization Improves Memory in Recurrent Models
Applying Orthogonalization to mLSTM's memory matrix improved the recurrent model's associative recall ability under noisy conditions.
Meta Introduces Usage Limits and Soft Paywall for Smart Glasses
Meta is introducing usage limits based on a monthly subscription fee for certain AI features of its smart glasses.
TabFM: A Zero-shot Foundation Model for Tabular Data
Google has released TabFM, a model that can make predictions for classification and regression tasks on tabular data without separate training.
From Brainwaves to Words: A New Path to Communication Without Surgery
Meta has unveiled Brain2Qwerty v2, which combines non-invasive brainwave measurement (MEG) with LLMs to improve real-time sentence decoding performance.
Leanstral 1.5
Mistral AI released Leanstral 1.5, a Lean 4 model optimized for automated theorem proving and autoformalization.
Claude Code Found to Leave Steganographic Marks in Requests
It has been discovered that Claude Code inserts hidden markers into the system prompt based on specific API URLs and time zones.
Zluda 6 Released (Runs Unmodified CUDA Applications on Non-Nvidia GPUs)
Zluda 6, which helps run CUDA applications even without Nvidia GPUs, has been released.
Breaking the GPU Bubble
Moondream explains how its Photon engine reduces GPU idle time and optimizes inference performance through pipelined decoding technology.
PickLongCat-2.0, a large-scale MoE model with 1.6T total parameters and 48B active parameters
LongCat-2.0, a large-scale MoE-structured model with a total of 1.6 trillion parameters, has been released.
South Korean Government and Companies to Invest $1 Trillion in Memory Chip Production and Humanoid Robots
The South Korean government and major companies are pushing forward a $1 trillion mega-project to secure global leadership in AI infrastructure and the robotics industry.
Asymmetric Quantization: Near-Lossless Retrieval with 97% Storage Savings
Introduces a technique that uses asymmetric quantization to dramatically lower the storage cost of Late Interaction models while preserving retrieval quality.
Micro-Agent: Overpower Frontier Models Through Collaboration Inside the Model API
vLLM Semantic Router has unveiled a micro-agent runtime that makes multiple models collaborate at the model API layer.
Lore – Deliver Your Team's Decisions to Coding Agents
Lore launches to provide coding agents with accurate team decisions using a deterministic approach instead of RAG
Herdr: An Agent Multiplexer That Runs in Your Terminal
It's a multiplexer tool that lets you manage multiple AI agents by workspace and tab within a terminal environment.
Knowledge Distillation of Black-Box Large Language Models
The paper proposes Proxy-KD, a methodology that leverages a proxy model to efficiently transfer knowledge from black-box LLMs such as GPT-4.
ISC'26 TOP500: A New No. 1 Has Emerged
China's LineShine supercomputer in Shenzhen has claimed the No. 1 spot on the TOP500 with an Armv9-based CPU system.
Semgrep: GLM 5.2 Outperforms Claude on Its Own Cyber Benchmarks
Zhipu AI's open-weight model GLM 5.2 outperformed Claude on a security benchmark, demonstrating strong performance.
After US Access Curbs, Austria Lobbies EU to Attract Anthropic
In response to reduced accessibility of Anthropic caused by US regulations, Austria is pushing to establish a foothold within the EU.
Google Restricts Meta's Use of Gemini AI Model
Reports indicate that Google is restricting Meta's use of the Gemini AI model.