Gemini’s ‘Personal Intelligence’ upgrade

Good morning, AI enthusiasts. As frontier models converge on capability, the differentiator is becoming personal context — and nobody has more of it across the web than Google.

The corporate’s latest ‘Personal Intelligence’ upgrade lets Gemini pull from Gmail, Photos, and YouTube routinely, turning the apps billions already use into an AI moat rivals will struggle to cross.

In today’s AI rundown:

• Gemini’s latest ‘Personal Intelligence’ upgrade

• McConaughey trademarks himself to fight deepfakes

• Get essentially the most out of Google’s Gemini in Gmail

• Z AI’s open image model trained on Huawei chips

• 4 latest AI tools, community workflows, and more

LATEST DEVELOPMENTS

GOOGLE

🧠 Gemini’s latest ‘Personal Intelligence’ upgrade

The Rundown: Google just launched Personal Intelligence, a brand new beta feature that lets Gemini reason across apps like Gmail, Photos, YouTube, and Search data to deliver more personalized responses without users needing to specify which app to drag from.

The small print:

• Personal Intelligence connects Google’s app suite to Gemini, letting the assistant understand, locate, and proactively use personalized details.

• The tool can reason across text, images, and videos, with VP Josh Woodward detailing the AI referencing his photos and emails to assist at a tire shop.

• Personal Intelligence is off by default, with Google saying it won’t train its AI models directly on connected info like inboxes or photo libraries.

• The feature is rolling out to Gemini AI Pro and Ultra subscribers within the U.S. first, with plans to expand to free tiers and AI Mode in the longer term.

Why it matters: This seems like Google is finally playing the cardboard its AI rivals will at all times struggle to match: billions of users already living inside Gmail, Photos, and YouTube. Because the frontier models all grow to be extremely capable for the common user, deep integrations with personal context from widely used apps can be the large differentiator.

TOGETHER WITH BOX

📦 Turn enterprise content into actionable data

The Rundown: Enterprises sit on gold mines of knowledge locked in content. Box Extract changes that, securely extracting structured data at scale to power faster decisions and automatic workflows.

Box is a game-changer for regulated industries, with use cases like:

• Financial services: extract loan terms and payment dates routinely

• Government: pull permit details and inspection dates to streamline compliance

• Insurance: distill claims info from reports and pictures to speed up processing

Learn more from Box.

AI & COPYRIGHT

⭐️ McConaughey trademarks himself to fight deepfakes

The Rundown: Oscar-winning actor Matthew McConaughey secured eight trademark approvals from the U.S. Patent and Trademark Office covering his voice, likeness, and video clips, in keeping with the WSJ, citing the necessity to combat AI deepfakes and misuse.

The small print:

• The approved trademarks include audio of his famous “Alright, alright, alright” catchphrase and short clips of him speaking and staring right into a camera.

• McConaughey told the WSJ that they “need to create a transparent perimeter around ownership with consent and attribution the norm in an AI world.”

• McConaughey’s lawyers say the filings give them a federal court avenue to pursue AI misuse, moderately than counting on state-by-state publicity laws.

• The actor is an investor in AI voice startup ElevenLabs, in addition to the face of Salesforce’s Agentforce TV campaigns.

Why it matters: The capabilities of AI image and video models are blurring reality greater than ever, and newer, powerful releases have been much more lax on creating real likenesses and IP. McConaughey is true to push against the unclear ownership rules within the AI era, but to date, it’s been a tricky legal game to play without strong results.

AI TRAINING

📧 Get essentially the most out of Google’s Gemini in Gmail

The Rundown: On this tutorial, you’ll learn how you can use Google’s latest Gemini features in Gmail, what is on the market at once, and essentially the most helpful and disappointing parts about this initial rollout.

Step-by-step:

1. With a Google Workspace or “AI Pro” personal account, go to gmail.com and click on on settings in the highest right. Be sure all of the “smart features” are enabled

2. In your inbox, click the “Ask Gemini” button next to settings and ask anything about your inbox, like: “What emails do I even have about [subject]?”

3. In any draft, you’ll be able to press Option + H (Mac) or Alt + H (Windows), or click “Help me write” to take Gemini’s help to jot down based on the e-mail’s context

4. After a reply is written, you’ll be able to click “Recreate” to regenerate it or click “Refine” to alter the tone or length of the reply.

Pro tip: Use “help me schedule” within the draft pane to have Gemini read the e-mail thread, check the calendars, and suggest meeting times to whomever you’re emailing.

PRESENTED BY SLACK FROM SALESFORCE

📈 The actual ROI of AI agents in collaboration

The Rundown: For all of the talk of AI’s transformative power, are firms actually seeing a tangible return? A brand new Metrigy global study of over 1,100 firms confirms that over 90% of organizations investing in AI are already achieving or expect positive ROI.

Research reveals that early adopters of agentic AI specifically are seeing:

• 21% reduction in operating costs

• 35% increase in customer satisfaction

• 31% improvement in worker efficiency

Download the free research report.

ZHIPU AI

🎇 Z AI’s open image model trained on Huawei chips

The Rundown: Chinese AI startup Zhipu AI just released GLM-Image, an open-source image generator hailed as the primary major model trained entirely on Huawei hardware — though early testing has not been positive despite strong benchmark scores.

The small print:

• The 16B-parameter model was developed using Huawei’s Ascend chips and software, with zero reliance on US semiconductors.

• Z AI claims GLM-Image excels at text-heavy images and beats Nano Banana Pro on accuracy benchmarks, though early user tests haven’t backed that up.

• The model trails top options like Nano Banana Pro and Seedream on overall image quality, but is released fully open-source under a permissive license.

Why it matters: Zhipu made its AI presence felt with a strong GLM-4.7 release in December (and a recent HK IPO), and now brings a capable, open image model debut trained entirely without Nvidia to the table. While it might not beat out closed rivals, it’s an indication that China’s AI industry is not waiting around for the chip war to finish.

QUICK HITS

🛠️ Trending AI Tools

• 👨‍💻 Thesys – Make AI apps respond with interactive UI as an alternative of partitions of text. Try one in every of the world’s first Generative UI API at no cost*

• 🧠 Personal Intelligence – Personalize Gemini with info from apps like Gmail

• 🎇 GLM-Image – Z AI’s latest open-source image generation model

• 🩻 MedGemma 1.5 – Google’s newly updated open medical models

*Sponsored Listing

📰 Every thing else in AI today

OpenAI announced a cope with chipmaker Cerebras to deploy 750MW of dedicated processing power for faster AI responses, with capability rolling out through 2028.

ElevenLabs is partnering with Deutsche Telekom to deploy AI voice agents for customer support, offering 24/7 support without wait times via app and phone.

Slack updated its Slackbot, now acting as a private AI agent that taps into messages, channels, and files to reply and leverage context throughout work.

OpenAI made its GPT-5.2-Codex model available to developers via the Responses API, extending the model’s availability beyond the corporate’s Codex platform.

Microsoft reportedly became one in every of Anthropic’s top customers, in keeping with The Information, ramping spending on its models to just about $500M annually.

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Jon M. in Wake Forest, NC:

“I coach youth basketball and at all times struggled with substitutions: how you can give every kid fair playing time while keeping balanced lineups on the court… I built an easy roster table, rated each player on height, offense, and defense (1–10), and asked ChatGPT to design a single-page HTML app that may generate substitutions.

The primary version worked, but I refined it using Claude for hands-on code edits, improving mobile usability and flow. Now I can toggle who’s present, select starters, and immediately get a quarter-by-quarter substitution plan with clear tracking.”

How do you utilize AI? Tell us here.

🎓 Highlights: News, Guides & Events

• Read our last AI newsletter: Meta’s massive AI compute push

• Read our last Tech newsletter: A moon hotel is taking reservations

• Read our last Robotics newsletter: Walmart expands drone empire

• Today’s AI tool guide: Get essentially the most out of Google’s Gemini in Gmail

• RSVP to next workshop @ 4PM EST Friday: AI foundations bootcamp pt. 2

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown