Functionalists point to MoE vs dense models to argue brain compute debate

JoshPurtell · x · 2026-10-06

In an X debate on consciousness and compute, @JoshPurtell laid out the functionalist position: functionalists consider the entire computational graph rather than a single metric. He argued that a 112B-A3B MoE model is generally stronger, more intelligent and more complex than a 3B dense model, and that the same logic extends to the brain. Acknowledging his earlier wording on complexity/compute was imprecise, he maintained that functionalists do not treat active compute as the dominant factor for consciousness. His interlocutor @VictorLevoso countered with physical limits: at 20W, the brain cannot exceed 10^21 operations per second under the Landauer principle, so any estimate of vastly more brain compute than current GPUs is impossible.

Original post →

Functionalists point to MoE vs dense models to argue brain compute debate

More from AGI Musings