Microsoft's New Tiny AI Model 🦾

Welcome to another edition of Horizon AI,

Microsoft has just released a new tiny AI model that matches the capabilities of models several times its size 🦾.

Let's get into it!

Read Time: 3.0 min

Here's what's new today in the Horizon AI

• A Visual to Kickstart Your Day 🖼: Lineup of humanoid robots from 2024

• Microsoft’s Phi-3 Shows The Surprising Power of Small AI Models 🦾

• AI Research: The World’s First AI-Created and Open-Source Gene Editor

• AI Tutorial: How to use Midjourney's new random style feature

• AI Tools to check out

• The Latest in AI and Tech 💡

A chart to kickstart your day

Lineup of humanoid robots from 2024

• The field of robotics is advancing rapidly. From Boston Dynamics' new all-electric Atlas to Elon Musk's Optimus Gen 2, each model brings something unique.

Source

AI News

MICROSOFT

Microsoft’s Phi-3 Shows The Surprising Power of Small AI Models 🦾

Microsoft Research announced Phi-3-mini today. This compact model punches well above its weight, delivering performance on par with much larger models.

Details:

• The Phi-3 model family offers three variants: Phi-3-mini (3.8B), Phi-3-small (7B), and Phi-3-medium (14B). Currently, only Phi-3-mini is available on Microsoft Azure AI Studio, Hugging Face, and Ollama, with the others set to arrive in the next few weeks.

• Phi-3-mini delivers advanced reasoning capabilities similar to much larger models like Mixtral 8x7B and GPT-3.5 at a significantly lower cost.

• The key to Phi 3's performance is the high quality of its training data set, which consists of meticulously filtered web and synthetic data, Microsoft says.

• Thanks to its compact size, Phi 3 can be run locally on a standard smartphone with as little as 1.8 GB of memory

The Phi-3 models represent the latest achievement in Microsoft's ongoing efforts to create high-quality, efficient, and cost-effective AI models. These efforts began last year with Phi-1 and have progressed through Phi-1.5 and Phi-2.

Source

AI RESEARCH

The World’s First AI-Created and Open-Source Gene Editor

A startup called Profluent is claiming to have created the world's first open source, AI-generated gene editor based on the CRISPR system.

Details:

• The researchers fed massive amounts of biological data into a large language model to generate new and more capable editors.

• The company claims to have already utilized one of these AI-generated editors, named OpenCRISPR-1, to successfully edit human DNA, although specific details are scarce.

• Profluent is open-sourcing the editor in hopes of leveraging the expertise of other researchers to improve the model over time.

• They aim to develop gene editors that are highly efficient and capable, and that will enable organisms to fend off diseases and other pathogens

The company has only just begun to explore the potential of AI-generated gene editors, and only time will tell if AI-created editors will emerge as a feasible treatment for humans.

Source

AI Tutorial

How to use Midjourney's new random style feature

Midjourney just released a new feature, called Style Random, which lets you “explore the latent space of visual styles”

To use it, just add ‘--sref random’ to the end of your prompt.

Essentially, it takes your prompt and explodes it into new and random style directions, broadening your creative horizons and potentially aiding in overcoming a creative block.

Example:

“A young woman with freckles wearing a yellow hat, close-up photo --sref random”

AI Tools to check out

💡 Fireflies: Helps your team transcribe, summarize, search, and analyze voice conversations.

🚀 Topview: Convert your ideas into viral videos automatically.

🖼 Fotor: Free online photo editor.

⚡ RapidLogo: Create your dream logo In seconds.

🌐 Collabwriting: Tool that helps you find, share and collaborate on stuff you find online.

The latest in AI and Tech

Microsoft expands its 'Content Integrity tools' to combat AI-powered election disinformation

Microsoft is expanding its Content Integrity Tools to combat AI-generated election disinformation. Offered to EU political parties, campaigns, and news outlets worldwide, these tools authenticate content origins, empowering organizations to educate voters against misleading information.

Coca-Cola to spend $1.1B to use Microsoft's cloud, AI services

Coca-Cola has inked a $1.1 billion five-year agreement with Microsoft to utilize its cloud computing and artificial intelligence offerings. As part of the deal, Coca-Cola and Microsoft will “jointly experiment” with Azure OpenAI, a service powered by OpenAI technology that enables customers to build chatbots and AI applications on Microsoft's Azure cloud platform.

Perplexity valued at $1 Billion, introduces a new enterprise offering

Perplexity has secured $62.7 million in new funding, nearly doubling its January 2024 valuation to around $1 billion. Additionally, the startup released "Enterprise Pro," a workplace search offering prioritizing security and control. This service intends to enhance search accuracy and efficiency for enterprise users.

OpenAI introduces ‘instruction hierarchy’ to protect LLMs from jailbreaks

OpenAI researchers have proposed an 'instruction hierarchy' for AI language models, aimed at reducing their susceptibility to prompt injection attacks and jailbreaks. This new approach shows promising initial results.

Tesla vehicles to get smarter with Grok AI integration

Elon Musk recently hinted at the integration of Grok, xAI's AI language model, into Tesla vehicles. This move could revolutionize how drivers interact with their cars.

That’s a wrap!

Thanks for sticking with us to the end! Let’s stay connected on LinkedIn and Twitter.

Not subscribed yet? Sign up here and send it to a colleague or friend!

See you in our next edition!

Gina 👩🏻‍💻