ImageBind 🤯

PLUS: GPT4 makes sense of GPT2

Hi folks!👋🏻 This is The Prompt! The latest and greatest in AI in under 5 minutes.

Let's get it

FEATURED

ImageBind: MultiModal Embedding by MetaAI 🔥

Meta AI unveiled their latest multi-modal embedding that works with text, images, videos but also depth, infrared, and the object's movement and position.

What are embeddings, again?

They help machines understand and make sense of complex data. For example, OpenAI’s text embeddings measure the similarity between text strings so you can use them in AI-powered search, grouping, recommendation, and even long-term memory.

ImageBind is a multi-modal embedding that can measure the similarity of all different types of data.

And that’s why this is so exciting.

What are the use cases for multi-modal embedding?

Endless possibilities. Here are some examples:

  1. Rich multimedia search - think multimedia Google AI search.

  2. Analyzing sentiment - looking at a person’s facial gesticulation + the words they say.

  3. E-commerce analysis on steroids - analyze the type of product images + buyer behavior data

Is it better than text embeddings?

Big yes.

For example, Multi-modal embeddings in natural language processing can consider the tone of voice and facial expressions, leading to a more complete understanding of the content.

Links

You can use this multimodal embedding API in ~30 lines of Python code!

Here’s the code & demo.

WHAT ELSE IS GOING ON

👧🏼 Wendy's X Google to develop an AI chatbot. This chatbot will take your drive-thru orders and can understand what “JBC” means. The plan is to launch the first “Wendy’s FreshAI” in Columbus, Ohio in June. Google is also partnering with other companies to help with AI.

👀 Google to announce PALM 2 & Bard on their event today. It’s rumored that they will announce multi-modal Bard and the second version of PALM that’s supposed to be good at coding & math. Here’s how to watch the event.

🏋🏻‍♀️  IBM shipped Watsonx, a new platform that delivers tools to build AI models. They were probably one of the early players in AI with Watson — years ago. Now, they’re back in the game. And they’ve partnered with HuggingSpace!

🏋🏻‍♀️  OpenAI wants AI to make sense of AI. They used GPT-4 to automatically propose explanations for the behavior of GPT-2's. The goal is to understand what happens under the hood & essentially understand the neurons work in the back of every AI. Mindblowing.

RESOURCES

The best resources we came across lately that will help you become better at writing prompts & building AI apps.

📚 13 use-cases for Code Interpreter [Twitter thread with examples]

👋🏻 Why AI Won't Cause Unemployment [a must-read]

🎥 What are vector & database embeddings [ Simple explanation ]

TOOLBOX

The latest AI tools to use or get inspiration from.

PROMPT OF THE DAY

TOOL

Midjourney

PROMPT

Output: Cinematic Shot Subject: French Supermodel, 18yo, tight evening dress, tall woman, beauty. Background: CocktailBar with red neon sigh lighting Influences: Red Lighting Emphasisers: RedCore, Red Color, Depth of Field Camera: Shot on a Hasselblad medium format camera. Carl Zeiss Distagon t* 15 mm f/ 2. 8 ze, Ricoh r1. --ar 21:9 --q 2 --v 5.1 --style raw

RESULT

What'd you think of today's edition?

Login or Subscribe to participate in polls.