Google announced that it is bringing Gemini Intelligence to Android, supporting cross-app tasks and widget creation.
Fine-tuning Qwen3.5-4B with RL for RLM achieved performance on par with Claude Sonnet 4.6.
Experiments with about 1,300 models revealed that the 20-tokens-per-parameter rule is tied to a specific tokenization method.
Introduced an Agent loop that automatically repairs outdated API/SDK examples using Codex.
Modal reduced serverless GPU startup to around 50 seconds using buffers and checkpoint restore.
AI data center demand is inheriting the EV/solar supply chains of power and analog semiconductors.
Google is discussing building orbital AI data centers with SpaceX.
Fast mode for Claude Opus 4.7 has been released as a research preview in the API and Claude Code.
Meta unveiled Muse Spark, strengthening Meta AI, Voice Mode, and AI glasses.
Vercel added Trusted Sources, which uses short-lived OIDC tokens for deployment protection.
Needle, an ultra-lightweight AI model with just 26M parameters distilled from Gemini 3.1, has been released.
Datasette 1.0a29 was released, including the addition of the `_r` utility and a segfault bug fix.
On TikTok, Mo Vitar satirized AI automation using the made-up term Ralph Loop.
Thinking Machines Lab has released a real-time interaction model for human-AI collaboration.
Mitchell Hashimoto pointed out the conservative decision-making tendencies of TDMs.
Google has announced Googlebook, a new laptop that evolves into an intelligence system by embedding Gemini AI at its core.
OpenAI summarized lessons on research challenges in the age of AI agents through Parameter Golf.
The **attention drift** phenomenon in speculative decoding drafters was identified and its performance was improved.
With Delta Channels in LangGraph 1.2, checkpoint storage for long-running agents has been reduced by more than 40x.
Vercel Firewall now lets you create custom WAF rules using natural language.
GitHub Copilot has released a preview of usage-based billing that calculates costs based on actual usage.
In Claude's next version, Fast mode will be applied as the default mode.
When running Claude Code, if an API key is present in environment variables, you may be billed for API costs instead of using your subscription plan.
Luce DFlash/PFlash boosts LLM inference performance by up to more than 3x on the AMD Strix Halo iGPU.