This AI model does it all

And Meta's new AI sunglasses

Happy Friday! I don’t know about you, but after watching the Rabbit R1 launch party keynote, I'm actually pretty excited to get my hands on one. 

Beyond the video/audio notes feature (and being able to order food from it!), I was interested in the R1 teaching mode. It turns the R1 into a handheld AI agent that learns tasks you show it and can perform them with just a voice command. Imagine a future where real-world actions can be voice-activated. 

Have you seen any reviews on the R1? Let me know your thoughts!

A “Jack of All Trades” AI Model Is Coming

Hugging Face

Picture an AI capable of complex vision, language, and decision-making tasks. Soon, this may be possible thanks to JAT: an AI project paving the future of AI versatility.

The details: Hugging Face, an open-source machine learning platform, just rolled out JAT—short for "Jack of All Trades.” The key idea? Instead of having separate AI systems for gaming, robotics, language understanding etc., JAT is a single neural network that could handle all of these different tasks.

  • When you give JAT a new task, whether it's a game to play, a robot to control, or a question to answer, it uses its broad knowledge from all that training data to try and figure out the best way to approach that task.

  • The JAT model was trained on the JAT dataset, a first-of-its-kind generalist AI agent training resource.

What’s possible with JAT? A generalist AI agent like JAT could complete a range of tasks, including:

  • Logistics coordination: Directing robots to streamline operations in warehouses.

  • Simulation assistance: Playing a crucial role in complex simulations for research and development.

  • Home automation: Seamlessly managing and integrating household AI systems, enhancing everyday convenience.

Can you use JAT? Yes…and no. Although it’s making leaps in AI agent capabilities, end users can’t quite get a JAT in a box. Actually running or querying the JAT model requires specialized computing infrastructure, coding skills to use the released source code, and the ability to set up the right environments/tasks for JAT to operate on.

But anyone can contribute to its development. As an open-source project, all the code, models, and datasets associated with JAT are publicly available. This allows other researchers and developers (with the right infrastructure) to improve upon, adapt, and deploy JAT in various settings.

Why it matters: Trained across a variety of tasks—from gaming to real-world navigation—JAT shows the shift towards generalist AI agents capable of tackling multiple challenges with a single skill set. 

We’re moving beyond simple chatbots, which means we’re a step towards deploying fully-equipped decision-making AI systems.

Microsoft's Small AI Models Have Big Potential

Microsoft's latest innovation in AI, the Phi-3 series, is small in stature but poised to make a massive impact. 

Meet the Phi-3 lineup:

  • Phi-3 Mini: A dynamo with 3.8 billion parameters capable of performing tasks on par with models ten times its size.

  • Phi-3 Small: Slated for release with 7 billion parameters, targeting more complex processing needs without the bulk.

  • Phi-3 Medium: A mid-range model with 14 billion parameters, designed to balance power and size for optimal performance.

But it’s more than just numbers. These models are redefining efficiency. The Phi-3 Mini, for example, compares favorably to larger models like the GPT-3.5, achieving similar capabilities but in a much smaller form. These streamlined models could eventually work directly on our devices.

The competition isn't far behind: Google’s Gemma and Anthropic’s Claude 3 Haiku have carved out niches in document summarization and dense academic research. But Microsoft’s Phi-3 still stands out beyond its size efficiency: These models prove wide-ranging utility—from simple concepts to complex reasoning.

Why it matters: This advance could significantly shift Microsoft's reliance on expansive AI models like those from OpenAI. Keep an eye on how these tiny titans might transform our everyday tech interactions.

Don’t Re-Explain—Create Visual Documentation

Avoid explaining the same information over and over by using Guidde’s generative AI platform to produce “how to” videos and documentation in seconds. Click capture on the browser extension and you’re good to go.

Guidde is great for:

  • Training and onboarding

  • New feature announcements

  • Getting started kits

  • Customer and employee FAQs

  • Integrations

The Guidde extension is 100% free—stop repeating yourself today.

Meta's Smart Glasses Just Got Smarter

June Wan/ZDNET

This Tuesday, Meta launched sweeping updates to its wearable tech—bringing multimodal AI capabilities to augmented eyewear. 

What's new with the Ray-Ban Meta?

  • Video Calling: Share your perspective in real-time via WhatsApp and Messenger. Whether it’s a sunset or your child’s first steps, you can now stream your view directly during a video call.

  • Meta AI with Vision: Just ask your glasses about anything you’re looking at, and get instant, relevant information or assistance. For example, while shopping, ask whether a pineapple is ripe and your glasses will help decide based on visual cues.

These glasses aren't just smart—they're a multi-functional assistant perched on your nose. With the latest updates, Meta AI can now deliver real-time translations of menus or signs you’re struggling to understand, making it an indispensable travel companion.

Why it matters: With all of the new AI-powered gadgets entering the scene, Meta’s Ray-Ban smart glasses aim to stand out by enhancing how we interact with our environment through vision. This move could be Meta’s answer to past criticisms of AI wearables (*cough* Humane’s Ai pin), offering a more seamless and integrated user experience. The question remains: Can Meta break through the “shiny new AI object” syndrome?

More important AI news: Dive deeper into this week’s hottest AI news stories (because yes, there are even more) in my latest YouTube video:

Are you ready for speech A.I.? Because it’s here, not in 6 months or a year, but now. On the latest episode of The Next Wave, we dive deep into the incredible potential and the inherent risks of this groundbreaking tech:

How do you pick the right AI for you? Let’s explore together.

And there you have it! Be sure to check out The Next Wave podcast! I can’t wait to hear what you think. Have a great weekend!

—Matt (FutureTools.io)

P.S. This newsletter is 100% written by a human. Okay, maybe 96%.