🥟 Chao-Down #229 Questions rise on the quality of Google's Gemini demo, McDonalds looks to apply generative AI to its operations, AI helps junior employees get ahead at major firms

Would you give an AI access to all your personal data if it could answer more questions about you? In a project codenamed “Project Ellmann”, Google engineers and product managers are proposing to use their latest Gemini AI to do exactly that.

The idea would be to use multimodal language models like Gemini to ingest a user’s search results, identify patterns in a user’s photos, and create a chatbot that can “answer previously impossible questions” about a person’s life.

Such questions could include: “When did I last see my siblings?”, “Where should I move to if I want to stay close my best friend?”, “Do I have a pet?” and many other deeply personal topics.

Ellmann would gather this context by using biographies, previous moments and the order photos were taken to be able to describe a user’s photos more deeply than “just pixels with labels and metadata”. All of this is enabled with the multimodal capabilities that Gemini showed off this week.

Google is one of the few companies that can offer this unique product experience with their AI models. Will they succeed?

-Alex, your resident Chaos Coordinator.

What happened in AI? 📰

Google’s best Gemini demo was faked (TechCrunch)

Google weighing 'Project Ellmann,' uses Gemini AI to tell life stories (CNBC)

McDonald’s says it will use Google AI to make sure your fries are fresh (The Verge)

Gmail’s AI-powered spam detection is its biggest security upgrade in years (Ars Technica)

Where are all the robot trucks? (The Verge)

AI Helps Junior Employees Get Ahead at Major Firms (Business Insider)

Always be Learnin’ 📕 📖

5 Challenges in Moving Upmarket in SaaS and How to Overcome Them (saastr.com)

📖 Data Engineering Design Patterns (DEDP)

How Companies Incentivize Layoffs—A Study of Corporate Career Incentives (scarletink.com)

Projects to Keep an Eye On 🛠

facebookresearch/Pearl: A Production-ready Reinforcement Learning AI Agent Library brought by the Applied Reinforcement Learning team at Meta. (Github)

OPTML-Group/Diffusion-MU-Attack: To Generate or Not? Safety-Driven Unlearned Diffusion Models Are Still Easy To Generate Unsafe Images ... For Now". This work introduces one fast and efficient attack methods to generate toxic content for safety-driven diffusion models. (Github)

PKU-YuanGroup/Video-LLaVA: Video-LLaVA: Learning United Visual Representation by Alignment Before Projection (Github)

The Latest in AI Research 💡

StyleAlign - Style Aligned Image Generation via Shared Attention (style-aligned-gen.github.io)

Qwen-Audio - Advancing Universal Audio Understanding via Unified Large-Scale Audio-Language Models (qwen-audio.github.io)

The World Outside of AI 🌎

Spotify’s not going for Pulitzers anymore (The Verge)A

The Marvels Box Office Won't Be Reported Anymore by Disney (qz.com)

Dying malls get second lives as lifestyle hubs (axios.com)

5 Mindsets That Drive Consumer Behavior (northwestern.edu)

‘Wobbly spacetime’ may help resolve contradictory physics theories (The Guardian)

One of 3 Gen Z Wants to Work for Themselves. Here's Why. (businessinsider.com)

Apprentice programs for office jobs are on the rise (qz.com)

One Last Bite 😋