TOGETHER WITH

Good morning, human brains, and welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

• Microsoft’s new AI models 🦾 🤖

  Microsoft unveiled Phi-3.5.

• Bot makes waffles 🤖 🍳 🧇

  Stardust Intelligence teased its Astribot S1. Again.

• Answering the web’s FAQs about AI 🎥 🙋‍♂️

  I answered your questions via livestream.

New here? Subscribe!😎

Peep today's ‘What Would You Do?’ at the bottom. 👇

MAIN COURSE

Microsoft’s new AI models 🦾 🤖

On Wednesday, Microsoft unveiled Phi-3.5. It’s a trio of compact AI models that outperforms much larger models.

What are the models?

• Phi-3.5 Mini: It beats Llama 3.1 8B in multilingual tasks.

• Phi-3.5 MoE: It surpasses Google's Gemini 1.5 Flash in reasoning.

• Phi-3.5 Vision: It matches GPT-4o in multi-frame visual tasks.

What's special about them?

All three models boast a 128,000 token context length, allowing them to handle long documents and complex conversations.

Who can use them?

They're available on Hugging Face under an open-source MIT license. Anyone can use, modify, or commercialize them.

Why do I care?

These efficient models could bring advanced AI capabilities to edge devices and offline environments, expanding AI's reach across industries.

How can I use them?

To use Phi-3.5 Mini, click here.

To use Phi-3.5 MoE, click here.

To use Phi-3.5 Vision, click here.

I’m confused, what is Phi?

In simple terms, they are small AI models that outperform Gemini Flash and similar models in reasoning capabilities despite their smaller size.

Tiny but mighty. 🐜 🤏

Back in May, we covered the Phi-3-vision and its Phi-3 model family.

SPONSORED BY MAMMOUTH

Access to all the best AI models for $10

Mammouth gives you access to the best LLMs:

• Claude 3.5 Sonnet

• GPT-4o

• Llama 3.1 405b

• Gemini Pro

• Mistral Large 2

Together with the best AI generated images:

• Midjourney

• DALL-E

• Stable Diffusion

and more for only $10.

SIDE SALAD

Bot makes waffles 🤖 🍳 🧇

On Monday, Stardust Intelligence teased its Astribot S1. It’s a humanoid robot with some serious party tricks.

What happened?

The Astribot S1 shows off a range of skills in a new demo video:

• Cooking: Whips up waffles from batter to plate

• Tea ceremony: Performs the delicate art of kung fu tea

• Music: Plays the yangqin, a Chinese hammered dulcimer

• Sports: Practices Wing Chun and shoots hoops

What does it look like?

It's got a human-like upper body on wheels, prioritizing dexterity over walking.

What’s under the hood?

Stardust's “Design for AI” architecture integrates AI with advanced robotics, allowing human-like learning and interaction.

When can I see it?

The Astribot S1 debuts at the World Robot Conference in Beijing on August 21. Commercialization is planned for later this year.

Am I supposed to be impressed?

While the demo is flashy, remember that it's in controlled conditions. Real-world performance remains to be seen.

I’ve never heard of this thing.

We covered it in more detail back in April.

SWEET DESSERT

Answering the web’s FAQs about AI 🎥 🙋‍♂️

Last night, I answered a subscriber’s question via livestream. If you have a question about AI, respond to this email. I might answer it in a future video.

What happened?

I asked various AI models, from ChatGPT to Claude to Grok, about God, heaven, and the nature of the divine. The results? Mind-blowing and wildly varied.

What did AI say?

• ChatGPT: played it safe, sticking to factual descriptions.

• Claude: got philosophical, asking follow-up questions.

• Grok: went full cosmic comedian, describing God as a "cosmic DJ."

Just text?

Nah. I had AI generate images of God and heaven and create gospel songs about unexpectedly meeting the divine in a grocery store.

Why do I care?

This deep dive shows how AI's “opinions” can vary wildly based on their training. It's a fascinating glimpse into how these models process complex, abstract concepts.

How do I watch it?

To watch it on Twitter (X), click here.

To watch it on YouTube, click here.

To watch it on Facebook, click here.

YOUR DAILY MUNCH

Cool Tool ⚒️

Reppert Factor

This is Bill:

Bill is the business owner of a small business that generates ten million dollars in revenue. Bill had his cousin Chad do all the IT work in his office.

Sure, it's messy, but it gets the job done.

Until today…

Today, Bill is explaining to his staff that Chad forgot to back up their networks. There was also a hack last night where their client credit card information was leaked.

All of this could have been prevented if Bill chose Reppert Factor to outsource his IT department.

So tell us… who's your Chad?

Startup News 💰

Google Research HeAR’s code. It’s an AI system designed to analyze audio, specifically coughs and breathing sounds, to assess health.

OpenAI announced a partnership with Condé Nast. This allows ChatGPT and SearchGPT, to display content from Vogue, The New Yorker, Wired, and more.

Research 👨‍🔬

MeshFormer — designed for high-quality 3D mesh generation using sparse-view RGB images and normal maps.

LongVILA — a cutting-edge visual-language model designed to handle long-context data, particularly for analyzing extended video sequences.

FURRY FRIENDS

Say “Hello” to Jaspurr. 🐈

Respond to this email with your pet’s name and pic for a chance to be featured here tomorrow!

MEMES FOR DESSERT

AI ART SHOW

WHAT WOULD YOU DO?

Land pets are boring… 🥱

Pick your supernatural marine creature? 👇

Ideas? Comments? Complaints?e

Respond to this email or hit me up on 𝕏.

Until next time 🤖😋🧠