Schmidhuber traces world models to his 2015 paper, Geiping cites 2012 SLIM NNs

jonasgeiping · x · 2026-09-03

In a discussion thread, Jürgen Schmidhuber argues that current world-model thinking essentially appeared in his 2015 paper On Learning to Think, which moved beyond the inefficient millisecond-scale planning of his 1990 neural world models toward planning and reasoning over abstract concepts. Jonas Geiping replies by pointing to Schmidhuber's 2012 paper Self-Delimiting Neural Networks, which introduces SLIM NNs — treating network weights as programs with a halt neuron, forming prefix codes, and enabling asymptotically optimal program search with online weight updates during activation spreading.

Related event: Schmidhuber Claims OpenAI's Recurrent Depth Traces Back to His 2015 Paper(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →