Now, build an "AI chip" that can never stop "thinking"; even when it's in "sleep mode" it "dreams". Always active. Run it with a monstrous amount of RAM and GPU arrays for active "thinking" and a long term memory store (massive SSD disk array) and use the "dream time" to sort, organize, and prune that memory. Put it into a hybrid constant training/inference state so it can do both/either; constantly updating it's own model weights. Maybe even a tool harness to allow it to update it's own software/request hardware upgrades? Maybe even give it tools to train and implement new/other model types it might need/want? How soon do we get real AGI/ASI? How soon after "The Matrix"/"Terminator"-like reality (or for that matter "Star Trek"-like reality if the AI decides on a benevolent existence)?