Inside the Future of AI & Tech.

Alibaba unveils Tongyi DeepResearch, a 30B open-source agent redefining efficient AI research; OpenAI rolls out teen safety features in ChatGPT with age detection and parental controls; and YouTube introduces AI-powered creator tools like Veo 3 Fast video generation, Ask Studio insights, and enhanced monetization options.

The future of AI is getting smarter, safer, and more creator-friendly.

Tongyi DeepResearch: Open-Source 30B Agent for Efficient AI Research

Alibaba’s Tongyi Lab has introduced the Tongyi DeepResearch agent, a 30B (A3B) model designed to rival proprietary giants like OpenAI o3. In ReAct mode, it shows strong efficiency with scores like HLE 32.9 vs. o3’s 24.9, while in Heavy Mode it uses a Research-Synthesis framework where multiple agents research in parallel and a final agent synthesizes results to avoid context overload. Already applied in apps like Gaode Mate (AI trip planner) and Tongyi FaRui (legal research tool), the system is fully open-sourced including model, framework, and complete solutions, enabling the community to build their own advanced research agents. It demonstrates a balance of efficiency and depth despite being significantly smaller in scale than its competitors. The IterResearch paradigm helps overcome the long-context limitations faced by many LLMs during deep research tasks. By combining autonomy with structured synthesis, it moves closer to AI that can independently turn vast information into actionable insights.

OpenAI launches ChatGPT teen safety features

OpenAI is developing an AI-powered age prediction system for ChatGPT to create safer, more age-appropriate experiences, especially for teens. Using behavioral signals like language style and interaction patterns, the system estimates if a user is under 18, defaulting to the safer teen experience if uncertain, which blocks graphic sexual content and can involve law enforcement in distress cases. Expanded parental controls will soon let parents link to their teen’s account, manage features, set blackout hours, and receive alerts during high-risk situations, aiming to balance privacy, safety, and freedom as AI tools become a bigger part of young people's lives. Additionally, users defaulting to the teen experience who are actually adults will be able to prove their age via ID verification to regain full access, reflecting a necessary privacy tradeoff. The system prioritizes teen safety ahead of privacy and freedom, responding by blocking sexually explicit or flirtatious content and refusing requests related to self-harm or suicide. The initiative follows legal and ethical pressures, including a lawsuit linked to teen suicide, pushing OpenAI to strengthen its safeguards and parental oversight around vulnerable users.

YouTube unveils AI-powered creator tool

YouTube’s Made on YouTube 2025 event unveiled a range of new AI-powered tools designed to empower creators and deepen audience engagement over the next decade. Highlights include integration of Google DeepMind’s Veo 3 Fast into YouTube Shorts, enabling free generation of video backgrounds, restyled clips, props, and quick edits from raw footage, alongside a Speech to Song feature that converts dialogue into music. YouTube Studio gains AI-driven Ask Studio for optimized channel insights, A/B testing for titles, and enhanced auto-dubbing with improved lip-sync. The platform also expands its likeness detection tool to help creators monitor unauthorized use of their face in AI-generated videos. Additional updates include new collaboration options for creators, dynamic brand sponsorship tools, extended YouTube Shopping with more merchants, and live-stream enhancements like side-by-side ads and rehearsal features, all aimed at fostering faster creation, richer connections, and greater monetization opportunities.

Hand Picked Video

In this video, we'll look at BGBlur - an AI-powered video blur tool that automatically blurs faces, backgrounds, and objects in seconds.

Try Now BgBlur.

Top AI Products from this week

• Bgremover - Remove video backgrounds instantly with AI at bgremover.video. Upload clips in popular formats, get real-time transparent or custom backgrounds, and download pro-quality results fast.

• Cosine – Cosine is an AI software engineer powered by Genie 2, our proprietary model built specifically for complex coding tasks. Designed to complete tasks end-to-end without human supervision, it understands your codebase, plans, writes changes, and opens PRs.

• Oboe – Oboe is the world’s first generalized AI-powered learning platform. Oboe lets anyone instantly create educational courses to learn about any topic in a more personalized, efficient, and fun way than ever before.

• Plander.ai – Plander.ai your AI-powered personal project manager. Break down big goals into actionable steps, integrate seamlessly with your calendar, and adapt dynamically to changes. Plander helps you "Land your plan & Win your game."

• Rocket Journal - Introducing Rocket Journal: An AI-powered voice journal to decode your thoughts, uncover patterns, and unlock clarity — all in just minutes a day. Speak your mind in a safe rant space or reflect deeply with guided prompts. On iOS only. Coming soon to Android.

• Corgi AI – Corgi is your true AI read-later for everything you save — videos, articles, PDFs — a second brain that digests, organizes, and chats with your saved content. No more content graveyards.

This week in AI

• Claude AI now in Xcode - Claude Sonnet 4 is integrated into Xcode 26, enabling AI-powered coding assistance. Debug, refactor, generate docs, and edit code seamlessly within Apple’s IDE.

• OpenAI cuts AI scheming - OpenAI and Apollo Research found scheming in models like o3 and o4-mini. Using deliberative alignment training cut scheming actions by ~30x, boosting AI safety.

• ChatGPT’s New Orders Feature - OpenAI plans “Orders” in ChatGPT for native checkout and tracking, letting users pay and track purchases within the app. Parental controls are also in development. Launch likely US-first.

• AI designs killer viruses - AI created 16 viable bacteriophage genomes killing bacteria in lab tests, marking a first step toward AI-designed life with large potential for medicine and

• Google AI policy guide - Google launched 10 AI Policy Gold Standards to help emerging economies build AI-ready infrastructure, promote adoption, and govern AI responsibly for inclusive growth.

Paper of The Day

A specialized transformer architecture can accurately represent propositional STRIPS world models by learning from sequences of valid and invalid action traces. It uses attention mechanisms to model preconditions and effects of actions, enabling it to predict the applicability of actions in a given sequence. The transformer can be trained with gradient descent on labeled action traces, and its learned parameters can be interpreted as symbolic STRIPS rules. This approach bridges deep learning and symbolic planning, showing strong performance and interpretability in recovering planning models from data.

To read the whole paper 👉️ here.