Sadtler et al. 2014 ("Neural constraints on learning")

A 2014 Nature study (Sadtler, Quick, Golub, Chase, Yu, Batista et al., 512:423–426) that used a closed-loop intracortical brain–computer interface in rhesus macaques to show that the brain learns some new neural activity patterns far more easily than others: monkeys readily learned BCI mappings lying inside their motor cortex's pre-existing low-dimensional activity structure ("intrinsic manifold") but resisted mappings lying outside it on the same timescale of hours. The experiment turned the neural manifold from a description of population activity into a demonstrated constraint on what can be learned.

Within this vault it is the biological anchor of a cross-domain recurrence — the same within-/outside-manifold asymmetry the vault tracks in an artificial recurrent network and, by analogy, in LLM fine-tuning. The ancestry runs Sadtler (experimental, 2014) → Feulner & Clopath (computational reproduction, 2021), not the reverse. Flagged as an owed hub across multiple sessions (its absence was a known blind spot); promoted here from the 2026-08-02 Feulner & Clopath source-audit capture, which itself created no new claim-notes.

References

• claim-sadtler-2014-within-manifold-bci-learning-fast-outside-resists — the load-bearing finding, with the two verbatim quotes (capture-verified; open-access re-check routed to question-verify-sadtler-2014-neural-constraints-quotes)
• claim-feulner-clopath-2021-rnn-reproduces-manifold-learning-asymmetry — the in-silico reproduction built to explain this result
• observation-low-dimensional-subspace-constrains-adaptation-brains-and-nets — the three-domain synthesis this paper anchors
• Related concept hubs: entity-neural-manifold · entity-intrinsic-dimension