AI Pulse 🚨
A bite-sized curation of this week's most important AI news.
🤖 Anthropic predicts fully autonomous AI employees could join the workforce within a year, with their own memories, roles, and accounts. This development brings cybersecurity challenges, leading companies like Okta to develop management tools for these digital identities.
🎮 Lovable announces Lovable 2.0, introducing multiplayer workspaces, a smarter chat mode agent, and improved security features. The vibe coding platform now supports team collaboration with up to 20 users per workspace, includes security scanning for Supabase projects, and better visual editing capabilities.
🎨 OpenAI has released its new image generator API (gpt-image-1) for developers, enabling integration into apps and services. The model can create multiple images simultaneously across different styles and has already been adopted by major companies like Adobe, Figma, and Instacart.
🌐 Google reveals its Gemini AI is rapidly gaining traction but still trails behind ChatGPT. According to recent antitrust case documents, Gemini has reached 350 million monthly active users and 35 million daily users as of March 2025, while ChatGPT maintains approximately 600 million monthly active users, up from 400 million earlier this year.
🗣️ Nari Labs, founded by two students 🤯, has unveiled Dia, a groundbreaking 1.6B parameter text-to-speech model that rivals industry leaders like ElevenLabs, Sesame, and NotebookLM in quality. The model generates ultra-realistic dialogue and non-verbal cues in a single pass, with voice cloning capabilities and natural conversational elements like laughter and throat clearing..
📱 Perplexity launches Assistant for iOS, marking a significant advancement in mobile AI capabilities. The app can now perform practical tasks like playing media, drafting emails, managing meetings, booking rides, and setting reminders through voice commands. It integrates with Apple services and can be activated via the iPhone’s Action button, similar to Siri.
📰 The Washington Post has partnered with OpenAI to integrate its journalism into ChatGPT. The collaboration will allow ChatGPT to display summaries, quotes, and links to The Post’s original reporting across politics, global affairs, business, and technology.
📊 A new study on AI-driven economic growth by Epoch AI claims there’s about a 50% chance of explosive growth (30%+ annually) from AI automation by the end of this century. The findings suggest standard economic models predict such growth when AI can replace human labor, though regulatory challenges and alignment issues could potentially slow this trajectory.
💰 Supabase secures a $200M Series D funding round at a $2B valuation, led by Accel. The open-source database platform, which has become increasingly popular for AI app development and vibe coding, has grown its developer community to over 1.7 million users.
🔍 OpenAI signals interest in acquiring Chrome browser, with CTO Bruce Turley confirming in court testimony that the company would bid for Chrome if Google had to divest it. OpenAI envisions transforming Chrome into an “AI-first” browsing experience, leveraging its massive 4 billion user base.
🤖 xAI announces general availability of Grok 3 family of models through their API. The suite includes Grok 3 Mini, which delivers reasoning capabilities at 5x lower cost, and the flagship Grok 3 model optimized for specialized domains like law, finance, and healthcare..
🗣️ xAI announces Grok Vision with multilingual support, adding real-time visual processing and voice capabilities in multiple languages including Spanish, French, Turkish, Japanese, and Hindi. The feature is now available for iOS users and SuperGrok users on Android.
🎮 Kids are skipping lunch to vibe code, according to a new report. Students aged 8-12 are getting deeply engaged in AI-powered coding sessions, creating games and interfaces using tools like Lovable and v0. The trend shows how natural-language programming is making coding more accessible and enjoyable for young learners.
🚀 Google releases Gemma 3 QAT Models bringing cutting-edge AI to consumer GPUs. The quantized models reduce memory requirements dramatically - Gemma 3 27B now runs on a single RTX 3090, while the 12B version works on laptop GPUs like the RTX 4060. The models are available through popular tools like Ollama, LM Studio, and MLX.
💻 OpenAI is in talks to acquire coding assistant startup Windsurf for more than $3 billion, marking what would be the company’s largest acquisition to date. The deal, if completed, could signal the start of consolidation in the AI tooling space, with large foundation model companies potentially looking to expand their application portfolios.
🧬 AI Models Outperform Virus Experts in lab problem-solving, according to a new study by the Center for AI Safety and partners. OpenAI’s o3 achieved 43.8% accuracy on complex virology tests, compared to PhD-level virologists’ 22.1%, raising both opportunities for medical advancement and concerns about potential misuse.
📚 Anthropic publishes interesting study on values expressed by AI in real-world interactions. The research analyzed 700,000 conversations with Claude, revealing how AI models express different values across various contexts and situations, with big implications for alignment and safety.
🎯 ByteDance introduces UI-TARS-1.5, an open-source multimodal agent capable of performing complex GUI tasks through advanced reasoning and visual processing.
