NVIDIA's 3D worlds

PLUS: Siri learns to swear

Good morning, human brains. Welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

  • LLM Minecraft Expert ⚒️

    The GPT-powered bot learns continuously without human intervention.

  • NVIDIA’s 3D worlds 🌎️ 

    Turning 2D videos into immaculate 3D models.

  • Siri learns to swear 🤬 

    No more ducking typos.

APPETIZER

GPT-4 takes on Minecraft ⚒️ 

AI researchers recently created a Minecraft bot using GPT-4 — and it’s scary good.

Voyager is what’s known as an embodied agent — an AI that can move and interact with its simulated or real environment, of its own volition.

It can accomplish tasks and explore within the game, much like a human player would, whether it’s building a base, mining iron, or building a Nether portal.

Unlike previous AI, Voyager starts with little training but has an understanding of the Minecraft world through an ‘automatic curriculum’. It uses the game state to make guesses at what would make the most sense to do at any given point.

As it learns new tasks, it adds them to its skill library. This library expands as the bot learns, and is extendable — meaning it can skill stack to perform more complex and novel behaviors over time.

Since you can’t exactly parachute GPT-4 into Minecraft and expect it to ASDW and left-click like the rest of us, the bot uses MineDojo (made by the same team behind Voyager). It’s an extra simulation layer on top of Minecraft, which lets the bot write and execute code to interact with the player.

Voyager’s traversal through the Minecraft tech tree

Voyager is more than 15x faster at advancing through the tech tree than its predecessors.

The team has a full write-up on the project here, and has released the source code behind Voyager on GitHub.

Our take: It’s impressive that LLMs like GPT-4 are enough to power independent learning behavior in open sandbox games like Minecraft. While the rules of Minecraft are bare-bones simple compared to our real-life environment, this experiment gives us a glimpse at what it might take to produce autonomous AI humanoids.

FROM OUR PARTNERS 

Apiary, an email inbox built for newsletters.

Newsletters are awesome! But why does the reading experience have to suck?

Not with Apiary. It keeps you engaged, organized, and on top of all your favorite emails.

With Apiary, you can:

  • View your inbox like a social media feed, not an Excel sheet.

  • Set reminders to never miss a new email again.

  • Take notes directly inside your favorite newsletter to maximize retention & capture all your ideas.

  • Stay organized with direct control over what emails enter your Apiary inbox.

For a limited time, Apiary is offering its lifetime Early Adopter Tier membership as a preorder for a flat price.

BUZZWORD OF THE DAY

NeRF

NeRF stands for Neural Radiance Fields.

It’s an AI computer vision technique used to create 3D representations using 2D images taken at different angles. By feeding them through a neural network, it’s able to take all the different angles from each image and stitch them together into a 3D model.

MAIN COURSE

NVIDIA’s AI creates 3D models from video

NVIDIA has announced a new AI model, dubbed Neuralangelo. It uses neural networks to synthesize accurate 3D models from 2D video.

Neuralangelo… like Michaelangelo. Ha, that’s a knee-slapper.

The model uses the same underlying tech as NVIDIA’s Instant NeRF, and NeRF has been around since 2020.

But compared to prior results, Neuralangelo seems to produce the highest quality outputs yet  all from video you could capture on any smartphone.

Generated drone shot ~NVIDIA

Past models have struggled to recreate fine details and textures. But here, NVIDIA has a trick up their sleeve.

The incredible level of detail is the result of a two-step process. First, the AI selects the best frames out of the input video to construct the 3D model. Then, much like a sculptor, it refines the render by sharpening the details.

Enhance!

At the end of the process, you have a full 3D scene or object ready to use in 3D art, video games, or as virtual twins in robotics and industrial contexts.

Our take: this may just be the best implementation of NeRF tech we’ve seen to date. Huge kudos to NVIDIA.

MEMES FOR DESSERT

YOUR DAILY MUNCH

Think Pieces

Too much AI: how VCs feel about big chunks of their portfolios pivoting to AI.

How AI-generated hate is rising online.

Startup News

Thanks to AI, iOS 17 will now learn your swears.

EV, AI, and semiconductors are still popping amongst VCs.

Research

Graphing neural networks to analyze molecular data.

From human-centered to social-centered AI: assessing ChatGPT's impact on broader society.

Tools

ConvoStack: a free, open-source chatbot widget & backend deployment framework for your website.

Feder: grow your Twitter audience with AI.

Outboundly: the Chrome extension for personalized outreach.

Tweet to Video: convert your tweets to TikToks, reels, and YouTube videos.

CaptionIt: create image captions instantly with AI.

CaptionIt

TWEET OF THE DAY

AI is replacing jobs at AI companies…

Tag us on Twitter @BotEatBrain for a chance to be featured here tomorrow.

AI ART-SHOW

Until next time 🤖😋🧠

What'd you think of today's newsletter?

Login or Subscribe to participate in polls.