AI News
Daily AI industry news — funding, products, policy, and major moves from global AI companies and startups.
Perplexity Personal Computer Launches for Mac to Scale Local AI Agents
Perplexity released Personal Computer for all Mac users. The tool uses 400 connectors to control local files and apps. Perplexity will repla
Unsloth and NVIDIA Use Double Buffering to Solve LLM Training Bottlenecks
Unsloth and NVIDIA introduced a double buffering strategy for LLM training. The technique hides data copy latency to improve speed on B200 G
Z-Anime Brings 6 Billion Parameters to 8GB VRAM GPUs
Z-Anime uses full fine-tuning to improve anime image generation. The 6 billion parameter model runs on hardware with 8GB VRAM. It replaces t
ZAYA1-8B Hits 91.9% on AIME '25 Using AMD Instinct MI300
Zyphra released ZAYA1-8B trained on AMD Instinct MI300 hardware. The model achieves a 91.9% score on the AIME '25 benchmark. New MoE++ archi
Open Generative AI Collapses 200 Models Into One MIT-Licensed Studio
Open Generative AI released a studio integrating over 200 AI models. The platform offers five specialized studios and removes content filter
Wiki Builder: The Claude Code Plugin That Eliminates Knowledge Base Setup Tax
Wiki Builder is an open-source plugin for the Claude Code environment. It automates the creation of structured Markdown knowledge bases. The
Spotify's New CLI Tool Lets Developers Generate Personal Podcasts
Spotify released a beta CLI tool for generating personal podcasts. The tool connects agents like OpenAI Codex and Claude Code to Spotify. Ge
Why Anthropic Skill Scanners Miss the .test.ts Security Gap
Anthropic Skill scanners fail to detect malicious code in .test.ts files. Snyk found severe security issues in 13.4% of analyzed ClawHub ski
The 45x Cost Gap Between Vision and API AI Agents
Vision-based AI agents consume 45 times more tokens than API-based agents. API agents complete tasks in 19.7 seconds compared to 1003 second
Platos Moves AI Agent Infrastructure From Managed Clouds to Local Control
Platos is an open-source runtime for self-hosting AI agents. The platform integrates MCP gateways for Claude and Cursor. It supports deploym
The Output-Competence Decoupling That Masks Corporate Skill Gaps
AI tools create a decoupling between professional output and actual skill. The rise of AI slop increases reading costs and hides systemic er
Flow Maps: Solving the Sampling Bottleneck in Diffusion Models
Flow Maps reduce the iterative computational burden of diffusion models. The technology shifts from predicting local tangents to mapping glo
Why Stanford is Trading Parameter Space for Output Space Analysis
Stanford researchers propose analyzing output space instead of parameters. The theory explains benign overfitting and the double descent phe
Barry Diller Warns AGI Control Outweighs Leadership Trust
Barry Diller discusses the inherent risks of AGI at a WSJ conference. He argues that technical unpredictability outweighs leadership moralit
Snap and Perplexity Terminate $400 Million AI Search Partnership
Snap and Perplexity have ended their $400 million AI search integration deal. The partnership failed to move beyond limited testing in early
Why Elon Musk Lost Control of OpenAI During Its For-Profit Pivot
Court records reveal Elon Musk sought total control of OpenAI in 2017. Internal conflicts over for-profit status led to Musk's 2018 departur
The $45 Billion Valuation Driving DeepSeek's First Funding Round
DeepSeek is seeking investment with a valuation up to $45 billion. The lab optimizes high-performance models for Huawei hardware. Open-weigh
The $299 Reachy Mini App Store That Democratizes Robotics
Hugging Face launched an app store for the $299 Reachy Mini robot. The platform uses AI agents to convert natural language into robot code.
Google AI Overviews Integrates Community Forums to Combat Hallucinations
Google AI Overviews now incorporates real-time discussions from forums and social media. The update aims to reduce AI hallucinations by grou
Xbox Halts Copilot Development Amid Major Executive Overhaul
Xbox is discontinuing its Copilot AI assistant for mobile and consoles. CEO Asha Sharma is restructuring leadership with new CoreAI appointm
Sulphur 2 Brings Uncensored Video Generation to Local GPUs
Sulphur 2 enables uncensored video generation using LTX 2.3. The model supports both text-to-video and image-to-video workflows. Users can r
Telus Deploys Tomato.ai for Real-Time Agent Accent Conversion
Telus is implementing AI technology to modify call center agent accents in real time. The system uses a pipeline of ASR and neural vocoders
The Meaning Distortion Hidden in LLM-Edited Text
LLMs converge diverse human writing styles into a statistical average. AI-edited text prioritizes formal metrics over personal voice and cla
Uncle Bob's Shift to CRAP Scores as AI Ends the Coding Era
Robert C. Martin declares the era of manual code writing is over. AI now handles implementation while developers manage CRAP scores. Softwar
Bun Experiments With AI-Driven Code Porting From Zig to Rust
Bun founder Jarred Sumner is testing a Zig to Rust port using AI agents. The process employs Claude and a bot to apply 300 specific mapping
Why Did Google Chrome Silently Install a 4GB Gemini Nano Model?
Google Chrome is silently deploying a 4GB Gemini Nano model to users. The installation occurs without consent based on hardware VRAM checks.
GLM-5V-Turbo Boosts VQA Response Speeds by 30%
GLM-5V-Turbo introduces a native foundation model for multimodal agents. The model achieves a 30% increase in VQA response speeds over prede
SAP Acquires Prior Labs to Solve the Structured Data AI Gap
SAP acquired German startup Prior Labs to enhance structured data AI. The company is restricting AI agent access to Joule and NemoClaw. This
CLI-Anything and the New Blind Spot in AI Agent Supply Chains
CLI-Anything enables AI agents to execute repository functions via CLI. Malicious SKILL.md files bypass traditional SAST and SCA security to
AI Inverse Laws: Why Developers Must Stop Anthropomorphizing LLMs
Experts propose three inverse laws to govern human interaction with AI. The framework warns against treating probabilistic models as sentien