Some of the biggest names in AI are working on world models, and this is going to impact the way you work.

Right now, AI translates the world into words. Some of the greatest minds in AI are working on building spatial AI, which is models that predict what happens next in physical reality: how objects move, how they collide, how they persist over time.

Spatial AI models learn from observation - watching video, digesting spatial inputs, of which there is not as many to train on as text for example - and they build representations of the physicial without the need for explicit programming, so instead of predicting the next word in a sentence, they predict what happens next in actual space.

For example, you can upload a design of a building, and the model will create a 3d version for you. Feed them site footage and they understand context, movement patterns, environmental factors.

This is already happening! It isn't abstract research. Some of the early adopter architects are already beta testing world models in their practices.

The Key Players

#1 Fei-Fei Li (often called "the Godmother of AI"), her company World Labs' recently announced their first commercial release. You can play with it here: create 3d worlds with a single prompt.​

Fei-Fei Li recently published an interesting essay, "From Words to Worlds: Spatial Intelligence is AI’s Next Frontier.” I highly recommend you read it.
​
In it, she talks about how spatial intelligence evolved before language and remains foundational to how we learn and interact. Current text-driven LLMs, however sophisticated, remain disconnected from physical reality - they can't properly drive cars, guide hospital robots, or truly understand what they're observing.

Her vision is to build spatial intelligence to further advance the fields of scientific research, healthcare, and education. Imagine teaching a class of kids about a molecule by creating a 3d world of one and allowing each and every child to enter the world and experience what it's like to be in a molecule!

If you're familiar with the concept of 'digital twins', where a digital version of a specific place or environment is created, with a flow of data for sensors allowing for remote monitoring or maintenance predictions, then you might get a little excited about the potential of World Models.

#2 Yann LeCun: left his job as Head of AI at Meta last week to work on - guess what? World models, but from a different perspective. His view is that world models needn't be generative but should learn through observation, grounded in machine learning principles.

#3 The Tech Giants: Google, Meta and OpenAI are all working on world models too. That brilliant Veo from Google probably wasn't built with social media in mind, but rather building on Demis Hassabis' view of 'we've got intelligence; now let's use it to solve the world's problems'. Many of those problems require intelligence that operates in three dimensions.

Want to know more about Fei Fei Li?

​Here's a great podcast from Bloomberg that dropped this week. An interview with Mishal Husain and Fei Fei Li.

You get a really good sense of the person she is, her back story, her vision and she makes a really good case for the need to celebrate and support teachers.

On this week's Chatting GPT 🎙️

Geoff Allen, CEO of Mersus Technologies, has been building VR training for pharma and manufacturing since 2015 – long before 'spatial AI' became the latest fixation. His company creates controller-free, hand-tracked training simulations that cut learning time by 75-80%, capturing real human motion in regulated clean rooms where a single airborne particle can scrap a $10 million solution.

Their Avatar Academy platform runs on Meta's Horizon OS, built in Unity, and will roll out global multiplayer training by Christmas. Apple's entry last year accelerated adoption drastically. Tune in to hear why Geoff says, "If Apple's betting on it, you have to keep an eye on it," when it comes to spatial technology.

​Listen on Spotify. Apple Podcasts

Tell us what you think of our podcast

We've just completed season 6 and we'd love your feedback to help us plan future seasons. Got 3 mins to tick a few boxes?

Copilot Masterclass - how to make your Copilot license deliver ROI

Last course of 2025, and you are invited! We'd love to see you join us for this 3 hour session in which you will come away with your Masters in Copilot Usage! Bag your place and mark your calendar.

That's all we have for you this time.
Hope you enjoy exploring the world of spatial AI.

See you next time!
​The Team at AI Institute