In this episode, we take a deep dive into the fascinating universe of World Models. We'll unpack how they differ from traditional, purely predictive models, and explore key characteristics of World Models and how they empower AI agents to revolutionize the AI landscape.

We'll delve into the limitations of passive observation and into the power of intervention, expose the pitfalls of curve fitting in machine learning, and explore the "shadow problem" that makes building World Models from data so challenging.

Later, we'll discuss the powerful role of inductive biases in making World Models tractable. We'll also unpack the core principles of cutting-edge research areas like causal AI, generative flow networks, and active inference, and explore why these advancements are crucial for achieving artificial general intelligence (AGI).

Finally, Andre will share his startup journey and explain why he believes "a world model is all you need.”

Andre Franca is the co-founder and CTO of connectedFlow, which is developing the next generation of AI co-pilots. His previous roles include the VP of R&D at causaLens, and executive director at Goldman Sachs. Andre received his PhD in theoretical physics from the University of Munich.

Topics:

- Guest introduction and background

- Understanding world models and how they differ from purely predictive models

- The history behind world models and early research 

- The key characteristics of world models

- How AI agents will take the AI world by storm 

- Deep dive into how world models differ from traditional predictive models

- Passive observation versus intervention in the context of world models

- The main disadvantages of curve fitting in machine learning 

- Why is it challenging to build World Models from data?

- How the analogy of 2D shadows and 3D objects relate to the problem of tractability in World Models

- What are inductive biases and how do they help in making World Models tractable

- Core principles and significance of causal AI, generative flow networks, and active inference

- Why are World Models essential for achieving artificial general intelligence (AGI)?

- Tell us about your startup and why a ​”world model is all you need”

Show Notes:

More about Andre Franca:

⁠https://www.linkedin.com/in/francaandre/⁠

Resources:

World Model paper by David Ha & Jurgen Smithhuber where they explore building generative neural network models of popular reinforcement learning environments

⁠https://arxiv.org/abs/1803.10122⁠

NeurIPS 2018 World Models workshop - Can agents learn inside of their own dreams?

⁠https://worldmodels.github.io/⁠  

Open AI GYM (A toolkit for developing and comparing reinforcement learning algorithms ) and Gymnasium

https://github.com/Farama-Foundation/Gymnasium

Atari Game on GYM Retro - a platform for reinforcement learning research on games including 70 Atari games and 30 Sega games and over 1,000 games across a variety of backing emulators

⁠https://openai.com/index/gym-retro/⁠

This episode was sponsored by:  

Ai+ Training ⁠https://aiplus.training/⁠ 

Home to hundreds of hours of on-demand, self-paced AI training, ODSC interviews, free webinars, and certifications in in-demand skills like LLMs and Prompt Engineering

And created in partnership with ODSC ⁠https://odsc.com/⁠ 

The Leading AI Training Conference, featuring expert-led, hands-on workshops, training sessions, and talks on cutting-edge AI topics and tools, from data science and machine learning to generative AI to LLMOps

Never miss an episode, subscribe now!