Anthropic released Claude for Creative Work along with 9 MCP-based connectors.
This analyzes the risk of 'agentic drift,' where AI agents behave differently from intended due to model updates and other factors.
An investigation found that Chrome distributes the Gemini Nano 4GB model without authorization.
vibevoice.cpp, a ggml-based inference engine that can run Microsoft VibeVoice without Python, has been released.
DeepSeek V4 Pro delivered GPT-5.2-level performance on agentic benchmarks while costing 17 times less.
The US Senate Judiciary Committee passed the GUARD Act, which mandates age verification and certain disclosures for AI chatbots.
Artificial Analysis ranked the anonymous model Peanut #8 on its Text-to-Image leaderboard.
An AI agent beta for Unity 6+ has been released as an editor feature and MCP server.
LLM editing changed the conclusions, stances, and voice of writing, distorting its meaning.
Codex CLI has introduced a `/goal` feature to prevent long-running AI agents from drifting off task.
The LLM plugin for Datasette, datasette-llm 0.1a7, adds the ability to set default options per model.
llm-echo 0.5a0, which provides a fake echo model for LLM, has been released.
The Claude Code system prompt revised its file-reading and REPL rules.
John Gruber reported that Y Combinator's stake in OpenAI is about 0.6%.
MTPLX, an engine that leverages MTP heads to boost LLM inference speed by up to 2.24x on Apple Silicon, has been released.
A plugin that turns code and knowledge bases into interactive knowledge graphs has been released.
vLLM has resolved the unimplemented TurboQuant error affecting Qwen 3.5+ models caused by an issue with Mamba layers.
OpenAI announced the European Youth Safety Blueprint and its first 12 EMEA Youth & Wellbeing Grant recipients.
Meta plans to launch a personalized AI agent that performs everyday tasks based on the new **Muse Spark AI** model before Q4 of this year.
Random cross-layer attention enables layer-wise KV cache sharing.
OpenAI has expanded ChatGPT ad purchasing through partners and a self-serve Ads Manager.
The White House is considering a government pre-screening system before AI models are released to the public.
LLMs don't just fix grammar—they distort meaning by shifting the argument and tone of writing.
By jointly training the tokenizer and the generative model, the approach achieved FID 1.48 on ImageNet 256x256.