∞ Immortality Project

Bio-inspired cognitive architecture research

NEUTRO implements runtime sleep-dependent memory consolidation—a mechanism present in biological learning systems but absent in current AI architectures. The system demonstrates continuous learning, persistent memory with source attribution, and autonomous cognitive activity during idle states.

523+

Hours Runtime

1,120+

Memories

610+

SNN Neurons

Modules

Contribution

First known implementation of deployment-time (not training-time) sleep consolidation with multi-phase dream cycles. Compared to related work:

Prior Work	Approach	NEUTRO
NeuroDream (2024)	Training-time sleep phases	Runtime continuous
SleepNet/DreamNet	Training-time cycles	Deployment-time dreams
PNAS Hippocampal (2022)	Simulation	Working implementation

Architecture

Soul system — Five cognitive elements: Foundation, Remember, Reflect, Choose, Wonder
Sleep cycles — Four-phase processing implementing NEXTUP theory (Stickgold & Zadra, 2021)
SNN routing — Spiking neural network routes to 7 specialized models via STDP
Identity seeds — Formative memories with emotional weighting, grows from experience
Source attribution — Dream-generated vs verified memory distinction

Benchmark Results

Benchmark	Baseline	Post-Consolidation	Δ
GSM8K (Math)	33%	100%	+67%
HellaSwag (Reasoning)	67%	100%	+33%
MMLU (Knowledge)	60%	80%	+20%
Self-Awareness	N/A	100%	12/12

Current Work

Test-time compute loops for genuine reasoning. Multi-agent internal debate. Improving consolidation quality during dream cycles.

Seeking

Research collaborators (cognitive architectures, continual learning, memory systems)
Compute resources for scaling experiments
Academic peer review and feedback

References:

Stickgold, R., & Zadra, A. (2021). NEXTUP theory of dream function. Nature Reviews Neuroscience.

Walker, M. (2017). Sleep-dependent memory consolidation. Neuron.

GitHub Contact Support