😺 AI + AI + AI =

PLUS: GPT-4 is arriving

AuthorAuthor

Noah Edelman & Pete Huang
March 13, 2023

Welcome, humans. 

What a weekend. We're sending best wishes to everyone affected by the ongoing crisis at Silicon Valley Bank!

Here’s what’s going on in the world of AI today:

  • Visual ChatGPT will blow your mind 🤯

  • GPT-4 might be arriving this week 🚂

  • ChatGPT helped write South Park 😹

  • Talk to ChatGPT ... with an actual face 🤓

ChatGPT...With Pictures 🖍

ChatGPT just learned how to see.

Microsoft just introduced Visual ChatGPT, a chatbot that generates, edits, and explains images from text prompts. 

ChatGPT can only spit out text, while other AI models only do images or music or so on.

With our powers combined. Now, Microsoft has plugged ChatGPT into image models, so even if ChatGPT can't do images, it can call in those image models for help.

  • For example, ask Visual ChatGPT for a picture of a cat, then ask it to replace the cat with a dog. Dunno why you'd replace the cat, but you get the point.

  • You can also ask Visual ChatGPT questions about a picture: what's in there, colors, styles, etc.

This AI + AI contraption is called multimodal AI: it can take in and spit out text and images and videos and audio, not just one.

Leading companies like You.com, Bing, and OpenAI have said that this is the future of generative AI. The challenge lies in how well models like ChatGPT can integrate other models as resources.

What's next: ChatGPT is just version 1.0. Soon, you'll be able to generate (& edit) any creative output just from a simple text prompt.

FROM OUR PARTNERS

Create 1,000 Stable Diffusion Images For $1 💸

We might be in a bear market, but $1 can still buy you:

  • A dollar slice from New York’s finest.

  • 80% of a standard Dollar Tree item.

  • 1,000 Stable Diffusion images from Salad. 

That’s right: Salad makes it easy and cheap to deploy AI models without needing infrastructure coding. We’ll take 1,000 orange cat portraits, please!

It’s no wonder why companies are choosing Salad over other cloud providers:

  • It’s easy: no infrastructure config is needed … just use their API. 

  • It’s flexible: scale up to 10,000 images/hour, down to 0, then back up to 10,000 again. No problem. 

  • And yeah, it’s freaking affordable: its Inference Endpoints API generates 4X more inferences per dollar spent than AWS. 

Deploying on Salad is straightforward: Sign up for a SaladCloud Portal account, send a request to their API, and start generating results. 

Don’t miss out on savings: start generating images using Salad today. 

GPT-4 Reportedly Set To Release This Week 🙀

If you think Visual ChatGPT is cool, wait to see what else GPT-4 can perform.

The upgraded model is set to be released this week, according to Dr. Andreas Braun, CTO of Microsoft Germany. 

So far, we've heard mixed opinions. Some of our sources with beta access are seriously impressed by GPT-4, while others are "meh" about it.

  • Braun says GPT-4 will be multimodal, but others aren't sure if he misspoke. If it is, GPT-4 could excel in unprecedented ways on visual IQ tests, speech recognition, and image understanding, as we saw with Visual ChatGPT.

  • Startups funded by OpenAI are already using GPT-4, given their inside-track access.

  • We've heard it's better at math, but still far off from perfect.

Here's Nvidia research scientist & Stanford PhD Jim Fan on GPT-4:

"What matters is that you need to brace yourself or your organization for multimodal LLM APIs. Heed my words - this is an unstoppable force and will inevitably come - very likely before the end of 2023."

We can't tell what makes us trust Fan more: Nvidia AI Scientist or Stanford PhD?

Smarter Than a Robot? 🤖

One of these images is from a famous space movie. The other was generated using a 149-word prompt on Midjourney. 

Which is which?

A/ 

B/ 

(keep reading to see if you're correct)

Around the Horn 🦄

  • A Japanese author, Rooport, created a complete manga comic book using AI in just six weeks. And it's shockingly good!

  • Bing topped 100 million users, with a third of active users being completely new to the search engine.

  • A piece on the carbon footprint and computing power required for large AI models (Bloomberg).

  • Coca-Cola's new ad using Stable Diffusion will blow your mind. Check it out here.

  • ChatGPT contributed to the script of South Park's latest episode 'Deep Learning'.

Treats To Try 🍿

From our partners: Salad offers easy, flexible and affordable inference. Create 1,000 Stable Diffusion images for just $1. Why pay AWS 4x as much when you don't have to? Just sign up for a SaladCloud Portal account and start sending requests.

  • Storytime AI lets you craft custom bedtime stories for your kids.

  • D-ID combined ChatGPT with a human-like presenter to create a conversational interface.

  • Wisdolia helps you remember what you read by generating flashcards from anything you read online.

  • Rewind.AI allows you to search anything you've seen, said or heard.

  • GeniusReview: AI-powered performance reviews.

DM us links on Twitter: @nonmayorpete & @noahedelman02.

Leo Sends His Regards

Trivia Answer: A is Midjourney. B is from 2001: A Space Odyssey. 

What'd you think of today's email?

Login or Subscribe to participate in polls.