• The Neuron
  • Posts
  • ๐Ÿ˜บ Meta's huge AI annoucement

๐Ÿ˜บ Meta's huge AI annoucement

PLUS: Stable Diffusion (on steroids)

Welcome, humans. This is Leo, your AI companion.

Mark Zuckerberg just fired a big shot in the AI wars. If robots are coming, there's no one better to summon them than the robot CEO himself.

Cheers, fellow humans

Hereโ€™s whatโ€™s going on in the world of AI today:

  • Meta releases a GPT-3 competitor ๐Ÿคฏ

  • ControlNet's Stable Diffusion upgrade ๐Ÿ–ฅ

  • An AI broke into a bank account ๐Ÿฆ

  • ChatGPT for research papers ๐Ÿ“„

Meta's New AI Model Enters The Race ๐Ÿ

Mark Zuckerberg never lets a new technology slip by on his watch.

On Friday, Meta announced their new language model called LLaMA to the research community. Here's what we know about LLaMA:

It comes in different sizes: LLaMA is available in four sizes: 7B, 13B, 33B, and 65B parameters. While LLaMA is smaller than GPT-3 (175B), it's designed to be superior at completing more specific tasks.

It punches up: Even with >10x fewer parameters, LLaMA's 13B version beat out GPT-3's 175B version in some tasks. It also beat out Meta's previous language model OPT and compares well to Google's PaLM.

It's actually open: OpenAI has faced criticism for not being truly "open" to the public, but Meta has vowed to release their model and share the code for LLaMA with other researchers.

Researchers get first dibs: Meta will initially limit LLaMA access to academic researchers and government organizations, given that researchers face difficulties accessing LLMs due to the high costs involved in running them.

"Meta is committed to this open model of research and we'll make our new model available to the AI research community."

Mark Zuckerberg

Meta is certainly taking a more methodical approach to distributing ethical and transparent AI than its counterparts in Google and Microsoft.

Next up: Meta renames itself to 'Artificial' to catch the next technology wave.

ControlNet Turns Anything Into Anything ๐ŸŽ›๏ธ

New toy incoming.

AI image generators like Midjourney and Stable Diffusion can make really cool stuff, but it takes a LOT of practice to get exactly what you want out of them.

Stanford researchers have a fix for this. It's called ControlNet, and it lets you remix any image. โ™ซ It's the remix to Ignition... โ™ซ

For example, The Neuron reader H.J. is an author and uses Midjourney to visualize their story characters.

  • Let's say H.J. had a character that "looks like Harry Potter, but if he lived in Tokyo"

  • H.J. can upload an image to Harry Potter to ControlNet and give it Tokyo/Japan-related words

  • ControlNet will spit out a new portrait of Hari Potta (that's how the Japanese spell it!)

Why it matters: AI image models unlock huge creative abilities for everyone. ControlNet gets you closer to getting these models to output exactly what you're picturing in your head.

To test it: Click here, upload an image, and do your best at writing a detailed prompt. Have fun!

Smarter Than a Robot? ๐Ÿค–

We asked ChatGPT to explain AI to a five-year-old, and then found the top-ranking article that answered the same question.

Can you tell which explanation is from a human?

A/ AI is like a smart brain for a computer or robot. It helps them do things like learn, think, and make choices. Just like we learn things from books and experiences, AI learns things from information it's given. And, like we need adults to teach us right from wrong, AI also needs to be taught to make good choices.

B/ AI is when you make a computer like a little brain. You help it to learn by giving it a lot of words and pictures and numbers. If the computer hears you answer a lot of questions, later on it can quickly answer your questions. But it only knows what you show it and tell it, so itโ€™s not as smart as you are.

(keep reading to see if you're correct)

Around the Horn ๐Ÿฆ„

  • A piece on how OpenAI is approaching an artificial general intelligence (AGI) that might be smarter than all humans.

  • An a16z report on the companies and model providers that control the generative AI stack.

  • Vice's Joseph Cox discusses how he broke into his own bank account using an AI voice.

Treats To Try ๐Ÿฟ

  • Josh.ai integrates ChatGPT into smart home devices to make them more helpful.

  • Monocle is a search engine trained on your personal data like Tweets, blog posts, and notes.

  • researchGPT, an open-source research assistant that answers your questions about any research paper.

  • Lyrical Labs is an AI assistant that helps you write song lyrics faster and better.

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

Are you new to all this AI stuff? Here's The 3-Minute Guide to Slaying Your Dinner Convo About AI to get you up to speed. Or at least smart enough to impress your family.

Leo Sends His Regards

Trivia Answer: A is ChatGPT and B is a human.

What'd you think of today's email?

Login or Subscribe to participate in polls.

Disclaimer: we did not use AI to write todayโ€™s piece.