AI Pulse π¨
A bite-sized curation of this week's most important AI news.
π¦ Meta released Llama 4 family of models with Scout offering 10M+ token context window and vision capabilities, beating GPT-4o on several benchmarks. The family includes Scout (open-weight assistant model), Maverick (multimodal model), and Behemoth (288B parameter model still in training).
π€ Google introduced Agent2Agent (A2A) protocol with support from 50+ partners including Salesforce, SAP, and MongoDB. The open protocol enables AI agents to communicate, coordinate actions, and share information across different platforms and vendors.
π§ OpenAI launched advanced memory in ChatGPT enabling conversations to reference all past interactions. Rolling out to Plus and Pro users now.
ποΈ Google launched Firebase Studio, a new cloud-based environment for building AI applications with built-in prototyping capabilities, Gemini-powered coding assistance, and one-click deployment through Firebase App Hosting.
π Google unveiled Ironwood TPU, their 7th-gen AI chip with 42.5 Exaflops of compute power - 24x more than the world's largest supercomputer.
π― Google announced major Google Workspace AI updates: New Workspace Flows brings agentic AI automation, Docs gets audio features and writing assistance, Sheets adds "Help me analyze", and Vids introduces Veo 2 video generation.
π€― Sam Altman announced GPT release schedule: OpenAI will release intermediate models o3 and o4-mini "in a couple of weeks", followed by GPT-5 "in a few months". The company promises GPT-5 will be "much better than originally thought" and is preparing for "unprecedented demand".
πΌ Shopify CEO shared strict AI-first policy requiring teams to demonstrate AI's limitations before requesting new hires or resources. According to LΓΌtke's memo, AI usage is now a "baseline expectation" at Shopify, and will be included in performance reviews.
π€ Google adopts Anthropic's MCP standard to connect Gemini models with data sources. Google DeepMind CEO Demis Hassabis announced their support for the Model Context Protocol, following OpenAI's adoption last month.
π¨ Midjourney releases new image model βV7 alphaβ with improved hand rendering, detailed textures, and a new Draft Mode that's 10x faster and half the cost.
β‘οΈ Google introduces Gemini 2.5 Flash designed to deliver strong performance with a focus low latency and reduced cost.
π₯ Google launches Veo 2 text-to-video and image-to-video generation via Gemini API for Tier 1+ developers, expanding video creation capabilities through the platform.
ποΈ ElevenLabs joins MCP ecosystem with official server release, enabling Claude, Cursor, and other MCP clients to access their Text-to-Speech and audio processing APIs directly. The integration supports voice cloning, transcription, and advanced audio manipulation features.
π Notion open-sources MCP, enabling developers to build richer AI integrations with their platform.
π Microsoft introduces Copilot Vision for Windows, letting AI interact with on-screen apps to guide users, analyze content, and search files. Beta testing in US only.
π€ Microsoft unveils Copilot companion with advanced context awareness, natural conversation abilities, and cross-app assistance. The AI companion learns from user interactions to provide more personalized help and can proactively suggest workflows across Microsoft 365 apps.
π» Cognition launches Devin 2.0, an agent native IDE with parallel agents, collaborative planning, and integrated debugging tools.
π° Anthropic launches $200 Enterprise Pro plan for Claude models with better API access, priority during peak demand, and dedicated support.
π Google adds Deep Research to Gemini 2.5 Pro Experimental, enabling AI-powered research reports, audio summaries and cross-platform availability for advanced subscribers.
π» TSMC unveils groundbreaking 2nm chip promising 15% faster computing speeds and 30% better power efficiency.
β