AI News
Daily AI industry news — funding, products, policy, and major moves from global AI companies and startups.
Why UC Berkeley CS Students Are Failing Despite AI Assistance
UC Berkeley reports a surge in CS course failure rates linked to AI use. Students relying on AI tools show a decline in fundamental math ski
Lovable Scales Google Cloud Usage 5x as Vibe-Coding Hits Fortune 500
Lovable expands its Google Cloud infrastructure 5x to support vibe-coding. The startup reached 400 million dollars in ARR with only 146 empl
Why Mathematicians Warn Against AI's Rapid Conquest of Formal Logic
Mathematicians warn that AI's rapid progress threatens conceptual understanding. The ability to find correct answers differs from intuitive
Elixir v1.20 Debuts Gradual Type Inference to Catch Errors Early
Elixir v1.20 introduces a new gradual type checking system. The update improves compilation speed via multi-core optimization. New dynamic t
Gemma 4's Encoder-Free Architecture Unlocks On-Device Multimodal AI
Google DeepMind released the Gemma 4 open model series for multimodal AI. The 12B model uses an encoder-free design to optimize on-device me
Why AethexAI Used 1.7B Parameter Models to Conquer African Voice AI
AethexAI raised $3 million to build specialized Voice AI for Africa and the Middle East. The Kora series uses 300M to 1.7B parameters to eli
Gooey Eliminates External Dependencies for Zig GPU UI Acceleration
Gooey provides GPU-accelerated UI for Zig across multiple platforms. The framework uses a zero-allocation strategy with 8-byte argument limi
The University of Toronto AI Worm That Hijacks Compute to Spread
University of Toronto researchers developed an adaptive AI worm. The malware hijacks infected device compute to fuel its spread. Static secu
Gemma 4 12B Brings Multimodal Reasoning to 16GB Laptops
Google released Gemma 4 12B as an open-weights multimodal model. The model runs locally on laptops with 16GB of VRAM or unified memory. A un
Google Search Console Adds AI Opt-Out to Satisfy UK CMA
Google added an AI opt-out toggle to Search Console following UK CMA requirements. Opting out of AI search results will not affect tradition
Microsoft IQ Collapses Four Data Sources Into One Agent Layer
Microsoft launched Microsoft IQ to unify business data for AI agents. The system integrates four context sources to eliminate data silos. Ne
Stanford Law Professors Preferred AI Tutors in 75% of Blind Tests
Stanford Law professors preferred AI tutors over humans in 75% of blind tests. AI responses showed significantly lower educational harm rate
GitHub Copilot Moves Beyond the Browser With New Desktop Agent App
GitHub launched a native desktop app to centralize agent-based development workflows. The tool integrates local, background, and cloud agent
Surface RTX Spark Runs 120B Parameter Models Locally to End GPU Bills
Microsoft unveiled the Surface RTX Spark to run 120B parameter models locally. The device replaces variable cloud token costs with a fixed h
Why Claude Code Now Uses Dynamic Workflows to Fight Agent Laziness
Anthropic introduced Dynamic Workflows for Claude Code to eliminate agent laziness. The new Opus 4.8 model maintains previous pricing while
Why MI300X Now Runs DeepSeek-V4-Flash at Half the H100 Cost
AMD MI300X overcomes FP8 dialect errors to run DeepSeek-V4-Flash. The hardware offers double the memory of H100 at half the list price. Soft
Kapa's Indexing Captioning Strategy Solves Multimodal RAG Cost Bloat
Kapa introduced indexing captioning to reduce multimodal RAG costs. The system converts images to text captions during the indexing phase. T
OpenAI Sites Collapsed Server Deployment Into a Single Prompt
OpenAI released Sites to deploy web apps via simple prompts. The tool integrates Cloudflare Workers, D1, and R2 for serverless hosting. AI a
Qwen3.7-Plus Cuts API Costs by 60% While Boosting Multimodal Power
Alibaba released Qwen3.7-Plus with multimodal capabilities and 60% lower costs. The model outperforms GPT-5.4 and Claude-Opus-4.6 on visual
Microsoft ACS: The Open Standard Fixing AI Agent Rogue Behavior
Microsoft released the Agent Control Specification to standardize AI agent governance. The open-source framework uses four interception poin
UC Berkeley Embraces AI While Other UC Campuses Resist
The University of California system lacks a unified AI adoption policy. UC Berkeley integrates AI for productivity while other campuses rest
Why Craft Now Outweighs Speed in AI Product Development
AI tools have made rapid product development a baseline requirement. Relying on AI defaults leads to generic and uncompetitive products. Tru
OpenAI Codex Evolves Into an Interactive Enterprise App Platform
OpenAI is transitioning Codex from a developer tool to an enterprise app engine. New features like Annotations and Sites enable non-coders t
Why the Public Market May Struggle With OpenAI and SpaceX Valuations
Anthropic has secretly filed for an IPO with a rumored valuation of $965 billion. NVIDIA and MiniMax are pushing agentic AI through new hard
Trump Signs Downsized AI Executive Order to Prioritize Market Speed
President Trump signed a revised and downsized AI Executive Order. The new order reduces regulatory burdens to increase corporate autonomy.
Microsoft ASSERT Turns Natural Language Policies Into AI Test Suites
Microsoft released ASSERT to automate AI behavior and regression testing. The framework converts natural language policies into quantitative
Microsoft Project Solara Swaps Windows for Android to Power AI Agents
Microsoft unveiled Project Solara as an Android-based platform for AI agents. The system replaces traditional apps with task-specific low-po
Perplexity's Hybrid Inference System Routes AI Tasks Between Local and Cloud
Perplexity AI introduced a hybrid inference orchestrator at Computex 2026. The system autonomously routes tasks between local hardware and c
Why Microsoft MXC Shifts AI Agent Security to the OS Kernel
Microsoft introduced MXC to provide OS-level sandboxing for AI agents. The system uses the Windows kernel to enforce strict access boundarie
Michael Burry Warns of AI Bubble Amid Anthropic's $965 Billion Valuation
Michael Burry warns that AI valuations for Anthropic and SpaceX are inflated. Anthropic reached a $965 billion valuation following the Claud