Jürgen Schmidhuber - Part 2: JEPA, the Road to AGI, and Who Really Invented Modern AI

In the second conversation, Jürgen Schmidhuber argues that the joint-embedding predictive approach named JEPA in 2022 belongs to the same family as his 1992 Predictability Maximization, and traces adversarial learning back to his 1990 world-model work and 1991 Predictability Minimization, curiosity-driven minimax games he regards as the origins of GANs. He names physical AI as the real frontier, since systems that excel on screen cannot yet use human tools, and reframes continual learning and test-time training through his 1991 fast-weight work, with detours through Solomonoff's universal prior, AIXI and the Gödel machine, ending on scientific attribution.