What comes next.
New releases, interesting ideas, and the developments that matter to builders.
The latest
100 articlesResearchers Extract Hidden Reasoning Traces From Proprietary LLM APIs
New research shows encrypted chain-of-thought blocks from frontier models can be recovered in plaintext by replaying them into weaker sibling models.
Read the news →Google Search Is Dying and AI Is Erasing the Web's Memory
A new The Walrus essay argues AI search is erasing the internet's collective memory, as Wikipedia traffic collapses, archives buckle, and the public record gets contaminated.
Read the news →Zuckerberg hits out at closed AI rivals as Meta doubles down on open models
Mark Zuckerberg has attacked closed AI rivals as Meta returns to open-weight models, arguing open source is the path to winning the AI race.
Read the news →Muse Glimmer: Meta open-sources a 30B agentic LLM tuned for local AI workflows
Meta Superintelligence Labs open-sources Muse Glimmer, a 30-billion-parameter agentic model designed for always-on local agent workflows on consumer hardware.
Read the news →Docker Sandboxes offer disposable microVM environments for AI agents
Docker launches Sandboxes, disposable microVM-isolated environments for AI coding agents like Claude Code and Codex that run unattended, safely.
Read the news →ChatGPT now blocks requests to copy an author's style
OpenAI's ChatGPT is refusing direct requests to mimic famous writers, but may still capture their broad qualities. The change could have real legal implications for AI writing tools.
Read the news →Cloudflare unveils Kitesurf, an agent-first browser on Workers
Cloudflare introduces Kitesurf, a stateless, agent-first web browser that runs in V8 isolates on Workers and is designed specifically for the Agentic Cloud.
Read the news →DeepSeek V4 Flash 0731 Tops ARC-AGI-2 at 61.4% for Pennies per Task
DeepSeek V4 Flash 0731 hits 89% on ARC-AGI-1 and 61.4% on ARC-AGI-2 at a fraction of a cent per task, shaking up the ARC Prize leaderboard.
Read the news →AMD snaps up Taalas, the startup etching AI weights into silicon
AMD snaps up Toronto startup Taalas, whose model-specific chips etch weights directly into silicon and reportedly hit nearly 17,000 tokens per second in early demos.
Read the news →Qwen3.8 Max tops Artificial Analysis agentic index as best overall model
Alibaba's Qwen3.8 Max is now ranked the best overall model on Artificial Analysis' agentic index, topping flagship rivals in agentic capability.
Read the news →Demis Hassabis steps up as Google DeepMind Chair, Jeff Dean departs
Sundar Pichai announces a major Google DeepMind reshuffle: Demis Hassabis becomes Chair and Alphabet Chief Scientist, Koray Kavukcuoglu takes over Gemini, and Jeff Dean departs after 27 years.
Read the news →Cloudflare open sources Cloudflare OS, an agent platform for the whole company
Cloudflare has open sourced Cloudflare OS, a platform that gives every employee an AI agent and workspace grounded in company context, skills, and internal systems.
Read the news →