AI is no longer just an emerging technology, it’s becoming part of everyday life, shaping how people learn, work, and connect across the globe.

• ChatGPT has reached an incredible milestone of 700M weekly active users, now touching nearly 10% of the world’s adult population. Its use spans from casual problem-solving to professional writing and coding, showing just how deeply it’s woven into daily routines.

• Google’s VaultGemma introduces a new standard for privacy-first AI. Built with differential privacy, this 1B parameter model protects sensitive data, making it a trusted choice for industries like healthcare, finance, and government.

• OpenAI’s GPT-5 Codex is redefining coding workflows. From quick fixes to autonomous debugging and refactoring, it acts like a true development partner, seamlessly working across IDEs, GitHub, and mobile.

Together, these breakthroughs highlight a turning point: AI is scaling globally, becoming more private, and proving its value as a reliable partner in real-world productivity.

How ChatGPT is Used Globally: Insights and Impact

ChatGPT has experienced extraordinary global adoption since its launch, reaching about 10% of the adult population by mid-2025, with 700 million weekly active users worldwide. While early users were predominantly male, this gender gap has nearly closed, and growth is fastest in lower-income countries. The majority of interactions—over 70%—are non-work related, focusing mainly on practical guidance, seeking information, and writing. Writing is the dominant work-related use, showcasing ChatGPT’s unique ability in generating and editing digital content beyond traditional search engines. Computer programming and self-expression form smaller segments of usage. ChatGPT’s economic value is largely tied to decision support, especially for users in knowledge-intensive professions, thereby enhancing productivity. The study used automated, privacy-preserving classification to analyze millions of conversations without exposing raw user data, offering comprehensive insights into demographic and usage trends.

The Most Advanced Privacy-First Language Model by Google

Google has introduced VaultGemma, a 1-billion-parameter large language model specifically designed with differential privacy to ensure sensitive data remains confidential during training. Built from scratch by Google Research and DeepMind, VaultGemma applies advanced mathematical techniques to add calibrated noise, preventing the model from memorizing or leaking personal or proprietary information, making it suitable for use in sensitive sectors like healthcare, finance, and government. The model not only sets new standards for privacy-preserving AI but is also open-source, allowing researchers and developers to experiment with privacy-focused machine learning directly, with performance comparable to older non-private models, all while maintaining strong privacy guarantees.

GPT-5 Now Fuels Codex Coding

OpenAI has launched major upgrades to Codex, now powered by the new GPT-5-Codex model, which is specifically trained for real-world software engineering and excels at both quick coding sessions and lengthy, autonomous development tasks. Integrated with ChatGPT accounts, Codex now lets developers work fluidly across terminals, IDEs, the web, GitHub, and even the iOS app, offering seamless transitions between local and cloud environments without losing context. The upgrade introduces faster response times for simple tasks, enhanced long-running code review and refactoring, dynamic adaptation of reasoning time based on task complexity, and significant accuracy improvements on code benchmarks. Codex now reliably catches bugs, supports collaborative work, and can independently refactor or debug code for hours, establishing itself as a dependable teammate for end-to-end coding workflows.

If you’re a streamer and want distraction-free videos, BG Blur lets you instantly blur backgrounds or faces with AI, keeping the focus on you and your content.

Try Now

Top AI Products from this week

• RizzCalc – RizzCalc is the first AI financial modeling add-on for Google Sheets, enabling users to build and edit models effortlessly using natural language. The name RizzCalc pays tribute to VisiCalc, the first spreadsheet built in 1979, which changed the world.

• HeyHelp – Save 1 hour a day by making Gmail work for you. HeyHelp is an AI assistant that auto-sorts emails, highlights priorities, drafts replies in your voice, blocks cold pitches, and learns your patterns. Works with OpenAI, Google, and Anthropic.

• Bazaar V3 – Bazaar turns screenshots of your app into motion graphic videos using prompts.

• Mureka Agent Studio - Mureka Agent Studio is a new feature that makes music creation accessible to everyone. Just describe an idea, a mood, or even a meme, and the agents handle everything from lyrics to composition, turning your thoughts into complete songs in seconds.

• Cognitora – Cognitora is the cloud platform built for executing AI-generated code. Unlike traditional container platforms, Cognitora leverages high-performance microVMs using Cloud Hypervisor and Firecracker to deliver secure, lightweight AI-native compute environments.

• Haystack – Haystack is an AI-native code review platform that makes PRs feel less like a jigsaw puzzle and more like a guided story.

This week in AI

• Figma 2025 Key Updates - Figma 2025 boosts workflows with design-to-Buzz sending, smarter team/community templates, editable Figma Make templates, plus new Apple UI kits for watchOS, macOS, and visionOS

• Roblox Moments & AI - Roblox Moments beta lets users 13+ create, edit, and share 30-sec gameplay clips. New AI tools generate 3D items, voice translation, and text-speech APIs enrich creator features

• Penske Sues Google - Penske Media, publisher of Rolling Stone, sues Google alleging AI Overviews steal articles, reduce site traffic, and harm revenue, accusing Google of abusing its search monopoly.

• SpamGPT Threat - SpamGPT is an AI-powered spam toolkit automating phishing campaigns with email server hacking, filter bypass, and AI-crafted content, threatening enterprise email security.

• Open Lovable AI - Open Lovable is an open-source AI web app builder that clones any website URL into an editable, working React app instantly. Built with Firecrawl, Groq, and e2b.

Paper of The Day

AMLNet is a knowledge-based multi-agent framework that generates realistic synthetic transactions for anti-money laundering (AML) research. It simulates over 1 million transactions reflecting placement, layering, and integration phases with 75% alignment to Australian AUSTRAC regulations. AMLNet’s detection unit uses ensemble machine learning models achieving an F1 score of 0.90, precision 0.84, and recall 0.97, proving robust in identifying suspicious activity. The system iteratively refines transaction generation based on detection feedback, supporting regulatory compliance and behavioral realism. It operates with sub-millisecond processing latency and demonstrates strong adaptability across external synthetic datasets. AMLNet is publicly released for research to advance AML detection and benchmarking.

To read the whole paper 👉️ here.