AI World Simulators Are The Next Big Thing 🌎 🤖

Welcome back to the AI For Humans Newsletter!

As we prepare for OpenAI’s 12 Days of Shipmas (this morning they announced 12 straight days of live streams showing off new product), we’ve been delighted by two big announcements in the ‘prompt-to-world’ space.

First, Dr. Fei-Fei Li (the pioneering AI scientist behind ImageNet and more) and her new company World Labs just came out of stealth to show off their new ‘image-to-world’ model which will allow you to take any image and turn it into an explorable environment.

We’ve seen a few companies try this before but the depth of what's possible in World Labs has us giddy with excitement. Not only can it create new environments but it can re-light those environments, add physics and all sorts of other stuff. This is a HUGE step towards AI-infused video game worlds and better simulating OUR world so AIs can learn better and faster. Def take a spin through the demo in the blog post.

And then Google DeepMind announced Genie 2, their ‘prompt-to-video-game’ engine which takes everything we found fascinating about Genie 1 and dialed it up to Genie 11 (we are sorry for this).

Genie 2 moves from simple 2D video game worlds to 3D worlds which, as you can imagine, leads to a much richer array of worlds to be simulated and experienced.

Why does all this matter?

Well, much like OpenAI initially said Sora wasn’t just a video tool but a ‘world simulator’. World simulations like Genie 2 or World Labs offer a sandbox where AI can learn, experiment, and grow in ways that mimic real-world interactions. These environments help AI develop deeper understanding of context, behavior, and decision-making without the risks or limitations of physical testing. For AI development, it’s like upgrading from watching videos to stepping into a fully immersive, interactive classroom—unlocking new possibilities for innovation in creativity, training, and problem-solving.

And, ultimately, might just make ‘prompt-to-video-game’ a reality.

Apologies for the newsletter coming in a day late but boy is there a lot on the horizon.

-Kevin & Gavin

We 💛 This - ElevenLabs Speech-To-Speech

It’s easier than ever before to create multi-character, single actor visual performances with tools like Runway’s Act-One and Hedra. But can a single actor drive multiple voice performances?

We have the technology, and it’s called Speech-To-Speech.

Now a year from it’s original release, the tech has gotten really good. Plus ElevenLabs’ growing library of voices makes it easier than ever before for you to deliver a performative take in your own voice, only for it to then be generated in an entirely different voice.

This support article has some great examples that’ll get your wheels turning.