The Neuron
Posts
😺 Meta's magical new model

😺 Meta's magical new model

PLUS: Google tries to flex

Pete Huang
April 06, 2023

Welcome, humans.

When will AI vs. AI debates be a thing? Sports competitions? Music recitals? Art competitions? It's all coming sooner than we think.

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

🖼️ Meta: "Cut it out"
💪 Google flexes in the chip game
🏌️‍♂️ AI at the Masters
🛡️ AI defends you against spam

Snip Snip ✂️

Meta just released Segment Anything, a new AI model that can recognize and "cut out" objects in images, even if it hasn't seen the object before.

It also comes with demo you can use, a nice surprise. Try it with your profile picture or a picture of an outfit you like.

Use cases: Segment Anything on its own is mostly good for things like background removal and making collages. But your AR/VR might use Segment Anything to isolate individual objects you're looking at.

And, c'mon. It's Meta. It's for sure going to be used for ads - e.g., showing you similar items in Facebook Marketplace based on what's in the listing you're looking at.

Giving back: Along with the model itself, Meta also open-sourced two massive datasets it used to train this model. One of them is 400x larger than the previous largest.

Hats off, Meta. This was a good one.

Create World-Class Designs In Minutes👇

Designing The Neuron was a pain at first. We went through a million cat logos and failed hilariously to create our own graphics.

We should’ve used Designs.AI.

Designs.AI is a suite of AI-powered online design tools that turns your brand into the next Apple or Starbucks. We’re not exaggerating. With their generative AI, your business can:

🍎 Generate over 10,000 variations for your new logo
🎥 Transform your content into powerful videos in minutes.
🤯 Write unforgettable marketing copy for users.

The best part: it’s a tool for complete beginners. No design experience necessary.

Everything within 2 minutes. Smarter, Faster, Easier.

Battle of Big Machines 💾

Moooommm, the chip designers are fighting againnnn!

On Tuesday, Google released a paper claiming their custom chip (called the TPU v4) is 1.7x faster and 1.9x more energy efficient than market leader Nvidia.

That was impressive for a few minutes, until people realized Google was comparing to a now-outdated chip called the Nvidia A100.

A day later, Nvidia released new chip test results, which showed that their new main chip (the H100) is 2x+ better than the A100.

(Conveniently, Google said they didn't compare to Nvidia's H100 because it came out after Google's chip.)

So now, we're not sure exactly who is better! We just know the chips are getting better 😎

Zoom out: Flexing aside, Google's mostly doing this to promote its Google Cloud offering, which is the only way you can use their TPUs anyways.

Meanwhile, an Nvidia H100 is the size of a half sheet of paper and runs $33,000 each 💰

Around the Horn 🦄

Stripe reported in its 2022 annual letter a record number of AI startups using its platform.
Alibaba is set to introduce their own ChatGPT competitors soon.
Daft Punk broke up partially out of fears about AI.
Meta says generative AI is, indeed, coming to its advertising products.
IBM is making AI commentary for the Masters golf tournament. It's not very good.

Treats To Try 🍿

Featured: Flexible, scalable, affordable inference. Get 4x more inferences per dollar with Salad’s Inference Endpoints API. Start generating results in 2 minutes with Salad.
Featured: Upload your PDF and start asking questions in seconds. Get citations and page numbers for every search. Free plan available. Sign up for Doclime.

DoNotPay launched a GPT email extension that trolls scam marketing emails and text messages.
Wisdom AI lets creators plugin a conversational chatbot to their online courses and websites.
Generate a hyper realistic background for any picture with AI Backdrop.
WhimsyWorks: AI-generated "choose your own adventure" stories for kids.
Speakable creates auto-graded speaking assignments in second languages.
What happens when you put this prompt in ChatGPT?