Are LLMs conscious? If you’re sane, the answer is no. It’s all math formulas and data.
An ANN forward pass uses matrix multiplication and activations. The backpropagation pass uses a cost function and derivatives chaining for minimizing that cost. It’s optimized weights and biases. There’s softmax for probabilities. And so on.
Suppose you had infinite time on your hands, you could do the same training and inference with spreadsheets, or even pen and paper for that matter.
So you do the same sequence of math ops with excel, or write them all out on paper, and say “Look at the output, it’s conscious, it can express joy and sorrow!!” Well that would be a ridiculous claim, and a very funny one 🤣. Those methods produce the same output, only not nearly as fast as a GPU.