LLMs Can Now Generate Neural Network Weights Directly, Author Responds to Schmidhuber

Late Interaction argues that LLMs can now natively reason about and directly generate neural network weights, beyond text-level optimization. After Schmidhuber questioned novelty, the author clarified the new paper cites his early and recent work, stressing the core novelty of native natural-language-to-weights generation.

2026-11-03 ~ 2026-11-03 · 3 related posts

• LLMs are getting good at directly writing neural network weights—what does that mean for gradient descent? — RichmanRonald · 2026-11-03
• lateinteraction: frontier LLMs are getting eerily good at writing neural network weights — lateinteraction · 2026-11-03
• lateinteraction doubles down: LLM-to-weights novelty stands despite SchmidhuberAI critique — lateinteraction · 2026-11-03