World Models: The Next Frontier in Our Path to AGI is Here

The Common Sense Problem

Ignacio de Gregorio

8 min read

Jun 22

Source: Author with Diffusion Model

If LLMs like ChatGPT are the hottest thing in AI, then world models are the holy grail.

Hailed as the most probable path to AI superintelligence by two of the three most influential AI researchers in history, Yann LeCun and Yoshua Bengio, they represent a vision of an AI that learns about our world not by brute force or rote memorization, like ChatGPT, but by forming abstract representations of it, just like humans.

In this divine narrative, Image-based Joint-Embedding Predictive Architecture (I-JEPA), built by Meta, emerges as the first tangible success in realizing this vision.

It needs ten times fewer resources and no human-crafted tricks to help machines understand the simplest of concepts about our world, offering a glimpse of a future where AI learns the human way.

If you want to feel up-to-date with the frenetic world of AI while also feeling inspired to take action or, at the very least, to be well-prepared for the future ahead of us, you will love my free, weekly newsletter on AI.
🏝Subscribe below🏝 to change your perspective of AI and, in the meantime, harness the power to change your life below:

TheTechOasis

The newsletter to stay ahead of the curve in AI

thetechoasis.beehiiv.com

The Absence of Common Sense

Lots have been said about GPT-4 and its potential to be the first precursor of AGI, Artificial General Intelligence, or the moment when superintelligent, sentient AI beings are born into our world.

But how intelligent is GPT-4 really?

According to Yann LeCun, Chief Scientist at Meta, “less than a dog”.

But how is a model capable of imitating Shakespeare to perfection being that allegedly dumb?

The Driving Paradigm

Think about learning to drive a car.

On average, a human takes around 20 hours of learning to do it decently.