[sylware]

I don't know if having a real-time ML/inference model (brain simulation with realtime input/output) is even remotely possible for a significantly big size.

Let alone those models be 'stable'.