AI Pulse 🚨
A bite-sized curation of this week's most important AI news.
🚀 NVIDIA GTC 2025 - Jensen Huang revealed Blackwell Ultra GB300 chip, Vera Rubin (coming 2026), personal AI supercomputers (DGX Spark and DGX Station), and GR00T N1, an open-source humanoid robot foundation model.
🎯 OpenAI launches o1-pro API - Their most expensive model offers better reasoning with 200K context, starting at $150/1M input tokens—10x more than regular o1 🤯.
🌐 Claude 3.7 now includes web search — Anthropic adds real-time search to Claude with source citations. Unfortunately, it’s only available for US users for now.
🌐 Chinese Baidu launches ERNIE 4.5 and X1 — New models outperform competitors at much lower cost, with ERNIE Bot now free for all users.
🚀 Claude 3.7 achieves 50% success on long tasks - METR research shows that AI task length doubles every 7 months, with Claude 3.7 now completing 1-hour tasks reliably. At this rate, AI could handle week-long tasks by 2027-2028.
🔬 Google’s AI solves 10-year superbug mystery in 48 hours - AI co-scientist independently proposed the correct hypothesis for antibiotic resistance mechanisms, matching researchers’ decade-long findings.
🎯 Mistral Small 3.1 outperforms Gemma 3 and GPT-4o Mini - Mistral AI’s new 24B parameter model features improved text performance, multimodal understanding, and a 128k token context window.
🎨 Google adds Canvas and Audio Overview to Gemini - New features include an interactive coding editor with real-time previews and AI-generated podcast-style discussions from documents.
🗣️ Sesame releases CSM-1B speech model — An open-source version of its near-human speech model that shocked everyone a few weeks ago with emotional intelligence.
💪 OpenAI building massive data center - First facility for $100B Project Stargate will house 400,000 NVIDIA AI chips, becoming one of the biggest AI computing clusters when completed in 2026.
⚠️ Anthropic’s Red Team warns of AI security risks - AI models are approaching expert-level knowledge in cybersecurity and biology, requiring stronger safeguards as capabilities advance.
✨ Adobe launches 10 AI agents for content creation - New agents help marketers scale content, find target audiences, and automate repetitive tasks.
🤔 Fei-Fei Li warns of AI risks - “Godmother of AI” co-led policy report calling for regulation that considers future risks and requires tech companies to disclose safety tests.
🤖 Google DeepMind showcases Gemini Robotics - Multimodal models enable robots to interact with the real world using advanced spatial understanding.
🧠 Google adds Mind Maps in NotebookLM - Interactive visual summaries help explore connections between ideas and provide tailored learning tools.
💻 Codeium launches “Windsurf Tab” - AI code prediction feature integrates terminal and clipboard for better context-aware suggestions.
