Researcher Argues Circuit View Dissolves Deep Learning Mysteries

Researcher will cb argues that well-regularized machine learning is essentially program search with a complexity prior, and that adopting a circuit view dissolves most of deep learning's apparent mysteries. He equates recursive generalization with grokking and contends backdoors do not emerge spontaneously without adversarial data.

2026-10-16 ~ 2026-10-16 · 4 related posts

• Interp researchers debate superposition: backdoors need adversarial data, won't emerge from training — willcb · 2026-10-16
• Circuit view of deep learning: ML is program search with an epiplexity prior — willcb · 2026-10-16
• will cb: the circuit view dissolves most of deep learning's supposed mystery — willcb · 2026-10-16
• Circuit view may dissolve much of deep learning's mystery, argues researcher thread — willcb · 2026-10-16