State of the art AGI implementation: 210M parameter LLM trained on a single RTX 4090 rented from Runpod :-)