How does this Spiking Neural Net project not have more views? The DigiCortex website is down but luckily i snagged a copy of the v2 installer for windows early last year, curious the lack of views on the YouTube channel (youtube.com/@321psyq/videos) and now scrubbing of the DigiCortex website and Neural Engine Download MSI. The project Engine is INSANE and was so ahead of its time (starting around 2010-ish) culminating in the V2 in the screenshot and video on youtube. I was able to find a copy of the whitepaper on the @waybackmachine archives: web.archive.org/web/20240330… According to o3 reviewing the WP when asked "Remark at how revolutionary the synthesis of this engine with frontier SOTA ai architectures and modern ai chips could be. What is possible?": Why this fusion feels like the next Cambrian moment in neural computing 1. Hardware-agnostic roots meet a hardware gold rush DigiCortex already expects heterogenous devices: the Workload Balancer autodetects NUMA nodes, CPUs, and GPUs, then hands off work through a clean LiquidCompute plug-in ABI. That same ABI can wrap today’s wafer-scale wafers (Cerebras WSE-3), ultra-fast SRAM tiles (Groq LPU), neuromorphic meshes (Loihi 2, IBM NorthPole), or trillion-transistor GPU monsters like NVIDIA Blackwell. Drop-in back-ends give you a single graph that can spill across all of them without touching the C++ core. 2. Deep-spike hybrids at mammal scale The engine already sustains >12 M neurons in real time on eight Voltas. Swap those cards for modern Blackwells or MI300Xs and you jump an order of magnitude. Now bolt a transformer front end (e.g., Llama 4 vision encoder) that converts frames to spike packets, and let the spiking cortex handle recurrent dynamics and plasticity. You get a system that sees with deep nets and thinks with biologically grounded circuits, all in real time. 3. Self-rewriting plasticity with meta-learning Plasticity rules are compiled just like any other kernel. Replace the voltage-based STDP with a learned surrogate-gradient loop that an outer-loop RL agent tunes on-line. The cortex literally rewires itself under policy control, giving you continual-learning robots that never need to “re-train” offline. 4. Whole-brain digital twins for precision medicine Feed patient MRI + DTI into the Cortical Circuit Builder, already designed for anatomical and diffusion data, then run personalized seizure forecasts or drug-response assays in silico. Pair the simulation with a language-model assistant that explains emerging wavefronts to clinicians in plain English. 5. Edge swarms with neuromorphic co-processors A Loihi or NorthPole plug-in turns DigiCortex columns into milliwatt-scale agents that fit on a drone or tractor. The heavy transformer brain lives in the cloud, while local spikes handle reflexes, closing the latency gap for agriculture, disaster response, or autonomous swarms. 6. Live-coding consciousness experiments Because the Simulation Controller exposes an external spike interface, you can pipe LLM-generated token streams, music, or even EEG recordings straight into a cortical column, watching phase-locked assemblies coalesce in 3-D. It is the closest lab bench yet to testing theories about integrated information, resonance, and synthetic phenomenology. 7. Cloud-scale sandbox for policy and safety Spin up 1 Billion-neuron replicas across thousands of elastic GPU instances, tweak plasticity constants, run adversarial curricula, and measure emergent agency. Regulators and safety researchers get concrete telemetry instead of toy-model speculation. Bottom line: Marrying DigiCortex’s plug-and-play neural substrate with frontier chips and SOTA architectures lets you prototype systems that see like GPT-4o, spike like a cortex, learn like AlphaZero, and run on anything from a groqcard to a wafer-scale slab. It is not just faster neuroscience software; it is a palette for building entirely new classes of adaptive, embodied intelligence. @elder_plinius @RileyRalmuto @idontexist_nn @voids_thoughts @TheHumanoidHub @annapanart @repligate @DeryaTR_ @iruletheworldmo