Anthropic details three infra bugs behind degraded Claude replies, OpenAI and NVIDIA plan 10GW GPU datacenters, and Grok 4 Fast cuts tokens 40% and cost 98%.
Three overlapping infrastructure bugs between August–September caused degraded Claude responses due to misrouted context windows, token corruption, and TPU miscompilation. All issues are fixed plus improved monitoring and evals now aim to prevent recurrence.
OpenAI and NVIDIA plan to deploy 10GW of AI datacenters using NVIDIA GPUs, with NVIDIA investing up to $100B as each phase rolls out. The first gigawatt launches in 2026 on the Vera Rubin platform to support next-gen AI models.
AI labs are investing heavily in RL environments to train more capable agents, fueling a startup boom led by firms like Mechanize and Prime Intellect, though experts debate if the technique will scale effectively with risks like reward hacking.
OpenAI is accelerating its robotics efforts by hiring experts to build and train humanoid systems, signaling a pivot toward physical AI to pursue AGI beyond large language models.
China has banned domestic tech firms from buying Nvidia AI chips, dealing a major blow to Nvidia’s already struggling presence in the region despite recent U.S. policy reversals aimed at reopening sales.
Nvidia and Intel are teaming up to create fused CPU-GPU chips targeting AMD’s dominance in integrated graphics, with Nvidia investing $5B and becoming a major Intel CPU customer while both eye broader markets like laptops and servers.
Grok 4 Fast offers frontier-level AI performance with 40% fewer tokens and a 98% cost drop, blending reasoning and non-reasoning modes in one model while ranking #1 in LMArena’s Search Arena.
Chrome introduces Gemini as an AI assistant to help with tasks across tabs, adds smarter AI search in the omnibox, and expands AI-powered safety tools, marking a shift to a more proactive and secure browsing experience.
Notion 3.0 introduces AI Agents that can autonomously handle tasks like creating docs, managing workflows, and searching tools. Agents are customizable and will soon scale into teams. New features include database row permissions and expanded integrations.
A framework for building and shipping MCP servers with TypeScript. Designed with DX in mind, it streamlines development and lowers the barrier to entry for anyone looking to create and deploy powerful tools on top of the Model Context Protocol ecosystem.
The world's first AI video model with HDR generation and intelligent reasoning. Transform text into stunning 4K videos with Ray 3 AI's Draft Mode and Hi-Fi Diffusion technology.
Select GPT-5 with Thinking in ChatGPT on web to toggle thinking time in the message composer.
Developers can connect their Claude account to Xcode and power coding intelligence features with Claude Sonnet 4.
Trigger a review by entering /review in the Codex CLI.
Tool helpers simplify tool creation and execution with automatic input validation and a tool runner for automated tool handling in conversations.
Whether you’re building with GitHub Copilot, agents, or any AI tool that speaks MCP, this is the place to find what you need.
You'll see updated naming across their sites and documentation, but your technical implementation stays the same.
Codex is used daily across numerous technical teams at OpenAI like Security, Product Engineering, Frontend, API, Infrastructure, and Performance Engineering.
FC-RewardBench evaluates reward models in tool-use scenarios, revealing current models’ weaknesses. New outcome-based models trained on open LLM data show up to 25% better task performance and enable more efficient fine-tuning.
Checkpoint-engine cut RL parameter updates for Kimi K2 from 10 mins to 20s via pipeline, buffer, and RDMA optimizations. Now open-sourced, it supports high-speed weight updates, fault recovery, and multi-checkpoint management across major LLMs.
A curated weekly roundup of AI news, tools, and breakthroughs to stay ahead of the curve in a fast-moving industry.
Curated by devs, for devs. Helping 400+ developers stay ahead of the AI curve.