Yes, but in this case, you can see what the model is doing, and it is running on your actual computer. Whereas a lot of LLM providers tend to run their models on their own server farms today, partly because it’s prohibitively expensive to run a big model on your machine (Deepseek’s famous R1 model needs at least a hundred GBs of VRAM, or about 20 GPUs) and partly so that they have more control over the thing.

AI isn’t a black box in the sense that it is a mystery machine that could do anything. It’s a black box in the sense that we don’t know exactly how it’s working, with which particular probability vector/tensor is responsible for what, though we have a fairly good general idea of what goes on.

It’s like a brain in that sense. We don’t know which exact nerve-circuits do what, but we have a fairly good general idea of how brains work. We don’t think that if we talk to someone, they’re transmitting everything you say to the hivemind, because brains can’t do that.