[LINK] Unifying Predictive Coding With Backpropagation

A short LINK post signal-boosting an lsusr LessWrong post (and the underlying paper) showing that predictive coding can approximate backpropagation along arbitrary computation graphs. Scott adds genuine explanatory value for a general audience — a clear, accessible account of what predictive coding is (the leading theory of how the brain works), what backpropagation is, why the brain can't do literal backprop (neurons only fire one way), and why an approximation was long suspected — plus a measured take (he's less excited than lsusr about 'fusing' the fields or neuromorphic hardware, but agrees it's evidence that artificial and natural intelligence are more similar than the 'brain does something fundamentally more advanced' camp assumes).

Quality 58 · Solid. Solid (58). More than a bare link — a clear, useful lay explanation of predictive coding vs backprop plus a measured assessment; short and fundamentally a signal-boost of others' work.

Claude’s paradigm shift 44 · Moderate. Moderate (44). Usefully frames a real result (brain/AI computational convergence) without original analysis.

Real-world impact 1 · Negligible. 1 — a link post; no real-world effect.