🎶 Gemini can now generate music

Good Morning! Google is adding AI music generation to Gemini, letting users create short songs from text, images, and videos, right inside the chat. Plus, you will learn how to instantly generate comprehensive research reports and presentations using Gemini Deep Research.

Plus, in today’s AI newsletter:

• Google Brings AI Music to Gemini

• OpenAI Launches Crypto Security Benchmark

• Real-Time Human Behavior Comes to AI

• How to Use Deep Research to Prep for Any Presentation Minutes

• 4 new AI tools worth trying

AI NEWS

🎶 Google Brings AI Music to Gemini

Google is rolling out beta access to its new music feature in the Gemini app, powered by DeepMind’s latest audio model, Lyria 3.

• Generates 30-second tracks from text prompts, images, or video references

• Supports instrumentals and full songs with auto-written lyrics

• Available globally (18+) in multiple languages including English, Hindi, and Japanese

• Also coming to YouTube Shorts via Dream Track, with AI-generated cover art for sharing

Google is turning Gemini into a creative studio, not just a chatbot. By baking music generation directly into chat, it lowers the barrier to creative expression, and tightens Google’s grip on the AI creator ecosystem.

AI NEWS

🔐 OpenAI Launches Crypto Security Benchmark

OpenAI released EVMbench, a tool that tests how well AI agents can detect, patch, or exploit smart contract vulnerabilities, right as AI-generated code comes under fire in DeFi.

• EVMbench is built from 120 real vulnerabilities across 40+ smart contract audits

• AI agents were better at exploiting bugs than fixing them, raising safety concerns

• A recent AI-generated bug drained ~$2.7M from Moonwell users despite passing an audit

• OpenAI’s latest coding agent showed big gains, but still struggles with full security coverage

As AI writes more on-chain code, weak security isn’t just a bug, it’s a financial risk. Tools like EVMbench hint at a future where AI must be audited as aggressively as smart contracts themselves.

AI MODELS

🎭 Real-Time Human Behavior Comes to AI

Tavus introduced Phoenix-4, the first real-time model that renders every pixel live, listens while you speak, and expresses emotion the way humans do in face-to-face conversations.

• Renders every frame in real time using a hybrid Gaussian–diffusion architecture

• Actively listens with context-aware gaze, nods, head movement, and micro-expressions

• Generates and controls 10+ emotional states that transition seamlessly

• Full-duplex behavior: hears, understands, and reacts while the user is still talking

Phoenix-4 isn’t just video generation, it’s a real-time human behavior engine. For healthcare, therapy, education, and any empathy-driven interaction, this marks a big step toward AI that doesn’t just talk to humans, but genuinely engages like one.

HOW TO AI

🗂️ How to Use Deep Research to Prep for Any Presentation in 15 Minutes Flat

In this tutorial, you will learn how to instantly generate comprehensive research reports and presentation outlines using Gemini Deep Research, an AI-powered agent that autonomously browses the web, analyzes data, and synthesizes complete, cited slide decks for you in minutes.

🧰 Who is This For

• Professionals and executives prepping for client pitches or strategy meetings

• Students and educators needing deep, cited literature reviews for class

• Startup founders building data-backed pitch decks

• Anyone who wants AI to handle hours of background research and slide structuring

STEP 1: Access Gemini Deep Research

Head to gemini.google.com and ensure you are using a Gemini Advanced account. In the chat interface, look for the "Tools" menu or the model dropdown and select Deep Research.

At the bottom, you’ll see an input box where you can type your presentation goal.

For example, you can write: “I am presenting on the future of autonomous vehicles. I need a comprehensive report on current sensor trends, market growth, and an outline for a 10-slide presentation with speaker notes.” 

Hit Enter, and Gemini will immediately start analyzing your request.

STEP 2: Review the Research Plan

Gemini automatically outlines exactly what topics it intends to search for before it begins gathering data. Once the plan is ready, the tool will display the steps it plans to take. Browse through the proposed research steps.

If you need it to focus on something specific, like open-source models rather than expensive enterprise software, click "Edit plan" to adjust the direction using natural language, then click "Start research."

STEP 3: Let Gemini Execute the Deep Research

Gemini autonomously searches the web, reads long articles, and gathers real data from multiple sources to fill your report with accurate, up-to-date content.

This process usually takes 5 to 15 minutes (you can leave the tab; it will notify you when it's done). Once complete, the tool will display a massive, highly structured report complete with inline citations, data tables, and a comprehensive summary of your topic.

STEP 4: Convert to Slides and Export

Once your report is ready, look toward the top of the screen and click Export to Docs to save the raw research.

To get your actual slides, simply type a follow-up prompt in the same chat: “Turn this entire report into a 10-slide presentation, including slide titles, bullet points, and speaker notes.” 

Your presentation outline will be ready instantly, complete with structured, data-backed content. If you have the new Gemini Canvas integration, you can even export this directly into Google Slides!

Microsoft's Project Silica team details its laser-modified glass storage tech, saying tests suggest that it can preserve information for at least 10,000 years

Fei-Fei Li's World Labs raised $1B from Autodesk, a16z, Nvidia, AMD, Sea, and others to build its world models for robotics, scientific discovery, and more.

Microsoft confirms a bug that let Microsoft 365 Copilot summarize confidential emails from Sent Items and Drafts folders, and deployed a fix.

Accenture told execs promotions would require “regular adoption” of AI and is tracking individual weekly logins to its AI tools for some senior staff.

🧑‍💻 Claude Sonnet 4.6: Opus-level performance for everyday work

🌎 Tiny Aya: Cohere's small, open-source model covering 70+ languages

🤖 Manus: AI agents you can use on Telegram that remember things over time

🎨 Qwen-Image-2.0: Alibaba’s all-in-one image generation and editing model

THAT’S IT FOR TODAY

Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!

See you tomorrow :)

- Dr. Alvaro Cintas