Preprint — Not peer reviewed | Patent application pending Page CORTEX: A Biologically-Inspired Neural Platform with Default Mode Network Consolidation and Neurochemical LLM Routing
Italo Luigi de Paoli
Zenodo (CERN European Organization for Nuclear Research) April 15, 2026 preprint DOI: 10.5281/zenodo.19589587 (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Preprint |
|---|---|
| Duration | 60 minutes of unsupervised DMN processing; overnight DMN operation |
| Measures | recognition correlation |
| Topics | Default mode network |
| Key points | The authors report that unsupervised DMN replay raised recognition correlation from 0.89 to 1.000 within 60 minutes, that overnight DMN operation with ambient music produced persistent elevated activity and gamma synchronization, that unprompted semantic output ('suonare', 'famigliola') emerged, and that novel patterns reached maximum correlation within 2 seconds after overnight consolidation. A neurochemical adaptive router is claimed to reduce LLM computational cost by 93.2%. |
Abstract
AbstractWe present CORTEX, a biologically-inspired neural platform implementing two mechanisms derived from neuroscience that have not previously been demonstrated in artificial systems: (1) a Default Mode Network (DMN) that consolidates learned patterns through unsupervised internal replay during idle periods, and (2) a neurochemical adaptive router that reduces Large Language Model computational cost by 93.2% through biologically-inspired parameter selection.Four experimental results are reported. First, recognition correlation increased from 0.89 to 1.000 after 60 minutes of unsupervised DMN processing — without external stimuli or additional training. Second, overnight DMN operation with ambient music produced persistent elevated network activity and gamma synchronization not previously observed. Third, spontaneous semantic output ("suonare", "famigliola") emerged without explicit programming during overnight DMN processing. Fourth, learning speed for novel patterns increased significantly after overnight DMN consolidation, with new patterns reaching maximum correlation within 2 seconds.Keywords: default mode network, memory consolidation, neuromorphic computing, LLM efficiency, brain-computer interface, sparse coding, neuromodulation, adaptive computation