Learning Without Weights
Alexander Parnell · VektorGeist · 2026-08-14 · preprint, not peer reviewed
What that distance does as a learner, with no trained parameters at all: 5–10× less data than a matched network, forgetting measured at exactly zero and provably impossible by construction, and next-frame prediction that beats a trained network with zero learned parameters — at a higher order of accuracy, second in the frame spacing against the network's first.
Read the paper — 10.5281/zenodo.21612831
Open access under CC-BY-4.0. The DOI above is a concept DOI: it always resolves to the newest version.
Nothing here has been reproduced independently, outside the project that produced it. Every paper says so. Every page of the record says so. That is the ceiling on all of it, and it does not move until someone outside repeats the measurements.