We’ve had several implementations even that were most of the way there just limited largely by hardware constraints. Megahal from the late 90s is pretty similar to modern LLMs, just drastically simplified.