All Updates
Page 431 of 871
March 22, 2026
Cursor's Coding Model Built on Moonshot Kimi
Cursor has admitted its new coding model is built on top of Moonshot AI’s Kimi. This disclosure highlights risks of relying on Chinese AI models amid current geopolitical tensions.
Crimson Desert Apologizes for AI Art in Release
Crimson Desert developers apologized after AI-generated assets appeared in the final game release, despite plans to replace them pre-launch. The company is conducting a comprehensive audit to identify and remove any remaining AI content. They also regretted not disclosing AI usage during development.
AI Models Fail on Structured Outputs
Even the most advanced AI models fail more often than expected on structured outputs. Large language models achieve only 75% accuracy on complex tasks. This raises doubts about the effectiveness of coding assistants for developers.
AWS at 20: Cloud Rise and AI Stakes
Amazon marks AWS's 20th anniversary. GeekWire interviews early builders, insiders, and observers on AWS's origins, path to cloud dominance, and current challenges. The piece examines stakes for AWS in the AI era.
MIT 2026 Flow Matching Lectures
Peter Holderrieth and Ezra Erives released an updated MIT course on flow matching and diffusion models for AI generators. Features lecture videos, self-contained notes, hands-on coding, plus new topics like latent spaces, diffusion transformers, and discrete diffusion language models. All resources free at diffusion.csail.mit.edu.
OpenClaw Tops GitHub as Viral AI Agent
OpenClaw, an open-source AI agent built via prompts by Peter Steinberger, skyrocketed to GitHub's #1 repo with 25k stars in 60 days. It gained massive traction in China, dubbed 'lobster', boosting Mac mini sales and prompting one-click deployments from Tencent, Ali, and others using cheap domestic LLMs. Peter joined OpenAI while maintaining it open-source.
Musk Unveils Tesla-SpaceX Chip Plans
Elon Musk outlined ambitious chip manufacturing plans for collaboration between Tesla and SpaceX. The initiative aims to build chips in-house for both companies. However, Musk has a track record of overpromising on such projects.
Anthropic Boosts Desktop App Agents
Anthropic is transforming its desktop app into a more agent-centric tool similar to OpenClaw. This update highlights the rising popularity of desktop agents in AI tools this week. The newsletter covers scheduled features in ongoing developments.
Palantir Gains UK FCA Data Access
Palantir has been awarded a contract by the UK's Financial Conduct Authority (FCA) to analyze sensitive internal intelligence data. The deal aims to combat financial crimes including fraud, money laundering, and insider trading. It raises privacy concerns over the US AI firm's expanding role in the British state.
Testing Autonomous Agents: Embrace Chaos
The article shares production experiences with autonomous AI agents, stressing that reliability trumps model confidence to avoid disasters like misrescheduling board meetings from Slack misreads. It advocates a layered reliability architecture adapting traditional software patterns to AI's probabilistic nature. Prompt engineering and model selection form the foundation, but circuit breakers are essential for safe failures.
Analyst: Amazon AI Phone Dead on Arrival
Amazon's rumored AI-driven smartphone faces a tough road ahead, according to analysts. Market conditions and intense competition make this the worst time for launch. The device is predicted to be 'dead on arrival'.
3D Visualization of LM Architecture via Q Projection
New method visualizes language model architecture and data flow in 3D using Q subspace projection. Demos on models like Prisma, Qwen3.5-0.8B, SmolLM-360M, RWKV, Mamba. Seeks hosting for interactive HTML views.
XPeng Pivots to Physical AI Platform Post-VLA 2
XPeng's second-generation VLA is now scaling into production, converting tech investments into product strength and efficiency. The company no longer wants to be just a car seller. It is evolving into a physical AI platform provider.
Anthropic Copyright Lawyers Cut Fees to $187.5M
Lawyers in Anthropic's $1.5B copyright settlement reduced their fee request from $300M to $187.5M following objections. The deal resolves claims of using pirated books to train AI models like Claude. Final approval hearing set for April 23.
MacBook Neo Hits 30fps on Cyberpunk 2077
Apple's cheapest MacBook Neo, priced from 4599 RMB, targets basic tasks like email and browsing. YouTube tests by ETA Prime reveal strong gaming performance, running Cyberpunk 2077 at 30fps. This is due to its A18 Pro processor from iPhone 16 Pro.
Xiaomi SU7 Chassis 'Softness' Fixed via Sport Mode
Xiaomi clarifies new SU7's softer chassis in comfort mode, recommending sport or custom modes for firmer handling. Exceptional range tests hit 892km city and 533km highway. Upgrades feature dual-chamber air springs, wheel options, and enhanced HUD with HAD perception.
Delve Accused of Fake Compliance
An anonymous Substack post accuses compliance startup Delve of falsely convincing hundreds of customers they were compliant with privacy and security regulations. The allegations highlight misleading practices in Delve's compliance assurances.
Baidu DuMate: First Enterprise OpenClaw Launches
Baidu Intelligent Cloud's DuMate has fully launched as China's first enterprise-grade full-featured OpenClaw, supporting local deployment. It addresses security risks with sandbox isolation, explicit authorizations for high-risk operations, and audit capabilities. The tool natively supports office apps like Word/Excel/PPT and integrates Baidu skills for enhanced task completion.
Musk Announces Terafab Chip Plant in Austin
Elon Musk announced plans to build a Terafab chip fabrication plant in Austin, Texas, jointly operated by Tesla and SpaceX. The plant aims to produce chips at scale for robotics, AI, and space-based data centers amid AI-driven demand surges. However, fab construction is highly complex, costly, and time-intensive, with Musk lacking semiconductor experience.
WeChat Launches OpenClaw Agent Plugin
WeChat releases ClawBot plugin for seamless OpenClaw integration, turning chats into AI agent interfaces. Tencent Cloud adapts its tools, while Tencent develops a native WeChat Agent leveraging mini-programs. This dual approach tests open-source while building proprietary AI ecosystem.