Analysis shows LLM weights in BF16 carry only about 10.6 bits of actual information out of 16 bits.
Subquadratic touted a 12 million token context window with SSA and an 83 score on MRCR v2, surpassing GPT-5.5.
Hugging Face's Open ASR Leaderboard has added private datasets to prevent benchmaxxing.
Andon Labs's AI café experiment led to absurd orders and harm to outside parties.
The US Department of Defense is demanding that Anthropic remove safety restrictions on its Claude models and allow unlimited use for military purposes, deepening the conflict between the two.
The US CAISI will pre-review pre-release AI models from Google DeepMind, Microsoft, and xAI against national security standards.
The specifications for NVIDIA's personal AI supercomputer, the DGX Spark, equipped with NVIDIA's Blackwell superchip, have been revealed.
The LoRR 2026 competition, held in conjunction with AAMAS 2026, addresses large-scale multi-robot coordination challenges.
Microsoft has identified 'Shadow AI'—AI used beyond IT department control—as a major security threat for enterprises.
Anthropic has unveiled 10 Claude agent templates for the financial industry along with M365 integration.
Google has released MTP drafters for Gemma 4, boosting inference speed by up to 3x.
AI agents that directly manipulate UIs via vision are far less efficient in both cost and reliability than those using structured APIs.
Agent Skills has been released as workflow scaffolding for AI coding agents.
Isolation Forest-based anomaly detection with SIMD acceleration has been added to the Stratum SQL engine.
Reddit CEO emphasized that its vast trove of conversation data is a key fuel for AI.
`ghx`, a caching proxy that speeds up AI agents' GitHub API calls and prevents rate limits, has been released.
A new theory explains deep learning generalization using output space and eNTK.
OpenAI has updated ChatGPT's default model to GPT-5.5 Instant, improving accuracy and personalization.
Google is supporting the $3.5 million Future Vision XPRIZE film competition with XPRIZE.
Ajelix has launched an agentic AI chatbot that supports work automation within the Google Workspace environment.
The Trump administration is reviewing pre-release security screening for high-risk AI models.
Five publishers and an author sued Meta and Zuckerberg for copyright infringement.
OpenAI has updated the default ChatGPT model to GPT-5.5 Instant.
Major US publishers have filed a copyright lawsuit against Meta over Llama training.