In practice it often depends on things like implementing your multiplication with a lookup table. I think the Landauer limit would matter if we did best effort computing and could harness the actual amount of determinism we can get from the substrate. We are far from that.
Should we optimize for a version of free energy minimization over entire architecture, or would that emerge over a simple individual feeding function and a centralized reward allocator? What are implications for human societies, or are these already approximating good solutions?
-
-
I also wonder how much we should rely on physics for tiling. In principle, we can arrange all tiles in a line and let them learn a routing protocol that tiles into a lattice of arbitrary dimensionality. Physical neighborhoods are inflexible but give us the first dims for free.
-
Indefinitely scalable dimensions *must* be aligned with physical dimensions, but one can also have finite dimensions until local memory runs out.
- 3 more replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.