Jürgen Schmidhuber argues prediction models alone are insufficient for AI, requiring active decision-making and reinforcement learning

🚨🚨🚨Jürgen Schmidhuber (@SchmidhuberAI ) on The Information Bottleneck podcast!!!

The man, the legend, one of the godfathers of modern AI.

His lab worked out a lot of what we're scaling today (LSTM, world models, Transformer variants, GAN-style setups) back when compute was ten million times more expensive.

We went back to the beginning, and what was special about that moment in time. We covered so many topics: why prediction isn't enough, and why acting in the world needs a real decision-making process. We get into compression from Kepler to Einstein, why we still don't have a robot that can do a plumber's job, the PMAX vs. JEPA exchange, how "to Schmidhuber" became a verb, what he thinks about the other godfathers of AI, and a special question from the audience :)

🎧 Link in comments. Part 2 soon.

🚨🚨🚨Jürgen Schmidhuber (@SchmidhuberAI ) on The Information Bottleneck podcast!!!

The man, the legend, one of the godfathers of modern AI.

His lab worked out a lot of what we're scaling today (LSTM, world models, Transformer variants, GAN-style setups) back when compute was ten million times more expensive.

We went back to the beginning, and what was special about that moment in time. We covered so many topics: why prediction isn't enough, and why acting in the world needs a real decision-making process. We get into compression from Kepler to Einstein, why we still don't have a robot that can do a plumber's job, the PMAX vs. JEPA exchange, how "to Schmidhuber" became a verb, what he thinks about the other godfathers of AI, and a special question from the audience :)

🎧 Link in comments. Part 2 soon.