Module 18 · Frontier

The Edge of AI Research

In this module, we will learn the ideas that are shaping the future of AI, from models that learn an internal picture of the world to systems that improve themselves.

By the end of this module, we will understand where AI research is heading next.

Lessons

1. 18.1 JEPA: LeCun's Vision for World Model AI: How humans and animals learn by observing the world · Yann LeCun's vision of autonomous machine intelligence · A simple everyday analogy to build the intuition · What does JEPA mean · What is an embedding or representation space · The problem with predicting raw pixels · The problem with contrastive methods · The core idea of JEPA · The building blocks of JEPA · The energy-based view in simple words · How I-JEPA works (for images) · V-JEPA and the world-model vision · When and why JEPA matters
2. 18.2 World Models: Teaching AI to Simulate Its Environment: What is an environment, a state, and an action · What is a World Model · The human analogy: imagining a move before making it · Why we need a World Model · How a World Model learns: predicting the next state · The latent state: compressing what we see · Imagining the future: rolling out without touching the real world · Dreamer-style agents that plan inside the model · World Models and predicting the future · World Models in the real world
3. 18.3 Recursive Self-Improvement: Can AI Improve Itself Indefinitely?: What is Recursive Self-Improvement? · Why does Recursive Self-Improvement matter? · How does an AI get better today? · How does Recursive Self-Improvement work? · A simple example with numbers · Two kinds of improvement · What exists in the real world today? · Intelligence explosion · Where it works well and where it fails · Keeping humans in the loop · Recursive Self-Improvement vs Normal Training

← Module 17: Production AI Infrastructure · Module 19: AI Engineering Career Prep →