Ideal LLM Interaction: Linear Input, Nonlinear Processing

fkasummer · x · 2026-07-16

The author argues that ideal human-AI interaction is: users provide near-linear input, the model performs deep nonlinear processing, and then users continue to guide the process in a linear, controllable manner.

He points out that the core feature of LLMs is that they 'leak' high-dimensional nonlinear capability; a good interface should convert this capability into a controllable and iterable user experience.

Original post →

More from AGI Musings

AGI Musings channel →