This week in AI, we see innovations making digital assistants proactive, empowering developers, and shaping AI laws:

🤖 ChatGPT Pulse
OpenAI’s new Pro feature delivers personalized morning briefs using chat history, memory, and connected apps. It provides quick visual cards with updates, reminders, and tips to start the day informed.

💻 Gemini CLI
Google’s open-source tool brings Gemini AI to the command line for coding, research, and task management. It supports up to 1 million tokens and integrates with popular IDEs for smooth developer workflows.

🏛️ Meta’s AI Super PAC
Meta launched a super PAC to influence state AI laws, promoting pro-tech policies to protect U.S. tech leadership amid many new AI-related bills.

From smarter assistants to developer tools and AI policy, these updates spotlight AI’s growing influence.

ChatGPT's Pulse Feature Learns About Your Life—And Acts On It

OpenAI has introduced ChatGPT Pulse, a new feature for Pro users that transforms ChatGPT into a proactive daily assistant by delivering personalized updates every morning based on users’ chat history, memory, feedback, and connected apps like Gmail or Google Calendar. Pulse creates scannable visual cards with relevant suggestions, daily briefs, reminders, meal ideas, and goal-oriented tips tailored to the user’s ongoing interests or recent topics, aiming to serve as a dynamic, context-aware morning briefing directly in the ChatGPT app. This preview feature allows users to curate their feed—select topics, give feedback, and manage app connections—making Pulse a more anticipatory and personalized experience compared to traditional reactive chatbots.

Gemini CLI: Open-Source AI Agent for Developers

Google has introduced Gemini CLI, an open-source AI agent that integrates the power of Gemini directly into the command line terminal. Designed for developers, Gemini CLI offers lightweight, prompt-driven access to the Gemini 2.5 Pro model, supporting a vast 1 million token context window. It excels not only at coding but also in a broad range of tasks including content generation, problem solving, deep research, and task management. With industry-leading usage limits—60 model requests per minute and 1,000 requests per day for free—developers can rely on Gemini CLI for uninterrupted productivity. Gemini CLI also integrates seamlessly with Gemini Code Assist, Google's AI coding assistant that supports multi-step coding tasks, code completion, debugging, and transformation across popular IDEs like VS Code, JetBrains, and Android Studio. This combined ecosystem enables developers to work efficiently with AI-assisted code understanding, file manipulation, and issue resolution, accessible through both IDEs and terminals, offering a powerful and flexible AI partner for software development workflows.

Meta Launches Super PAC to Shape AI Laws

Meta has launched a new super PAC called the American Technology Excellence Project to combat what it views as overly burdensome AI and tech regulations at the state level. This political action committee aims to support state candidates who are allies of the U.S. technology sector and advocate for AI development, aiming to maintain American tech leadership amid a patchwork of over 1,100 AI-related bills proposed nationwide in 2025. Managed by bipartisan firms and led by Republican strategist Brian Baker, the super PAC plans to promote pro-technology policies and empower parents on AI-related online experiences. This move signals Meta’s intensified efforts to influence state-level AI laws, which many tech companies see as a more critical battleground than federal regulation. The initiative comes amid growing concerns about child safety and AI misuse but also raises questions about Big Tech’s increasing political influence in shaping AI oversight.

Hand Picked Video

In this video, we'll look at the new update by Claude, where they revealed the system prompt and the artifact update.

Top AI Products from this week

• Scrumball - Launch influencer campaigns within minutes with your AI team. Lillian, Hunter & Charlie handle brand analysis, creator discovery, outreach automation, and ROI tracking from 120M+ database. We manage everything, so you focus on growth.

• Fakeradar – FakeRadar.io protects your sensitive data by instantly detecting whether you’re speaking with a real person or a deepfake — in real time, with just one click. Stay confident in every conversation and be sure the person on screen is truly who they claim to be.

• Open Notebook – All the familiar UI of NotebookLM but with fully customizable apps like video generators, podcast creators, and more. Remix existing apps or create your own.

• Vibes - Vibes is designed to make it easier to find creative inspiration and experiment with Meta AI’s media tools. As you browse, you’ll see a range of AI-generated videos from creators and communities.

• Figma MCP - Figma's MCP server now has remote access, making your design context portable. Bring live design system and layout info directly into your IDE and AI agents. This update also adds support for Figma Make and improves Code Connect.

This week in AI

• Kling AI 2.5 Turbo Upgrade - Kling AI 2.5 Turbo boosts video creation with improved motion, stability, and prompt fidelity, delivering cinematic 1080p outputs faster and more reliably at a lower cost.

• Anthropic Models in Microsoft Copilot - Anthropic's Claude Sonnet 4 and Opus 4.1 models are now in Microsoft Copilot Studio alongside OpenAI. Admins enable access; choose models per task for flexible AI agents

• OpenAI GDPval AI Benchmark - OpenAI's GDPval benchmarks AI on 1,320 real-world tasks across 44 jobs tied to top US GDP sectors, showing frontier models nearing expert quality in work deliverables

• OpenAI GDPval AI Benchmark - OpenAI's GDPval benchmarks AI on 1,320 real-world tasks across 44 jobs tied to top US GDP sectors, showing frontier models nearing expert quality in work deliverables

• Meta's AI Regulation Fight - Meta invests tens of millions in a bipartisan super PAC backing pro-AI candidates to fight restrictive state AI laws and uphold U.S. tech leadership amid rising regulation

Paper of The Day

Large models like GPT-5 may achieve top scores on medical benchmarks, but stress tests reveal major flaws. They often exploit shortcuts, change answers with minor prompt tweaks, and generate confident yet flawed reasoning. An evaluation of six leading models across six benchmarks shows that high leaderboard results mask brittleness and inconsistent measures of true medical understanding. Benchmark wins do not signal real-world readiness—robust, clinician-driven evaluation is essential for trustworthy medical AI.

To read the whole paper 👉️ here.