ELI5: In neural networks we could take data numbers, multiply by a constant, add a constant, add them to each other, and then do the same to the results again a

ELI5: In neural networks we could take data numbers, multiply by a constant, add a constant, add them to each other, and then do the same to the results again and again. And you could learn the right constants using labeled data. But we couldn’t multiply data numbers against each other. So in transformers you can. But you must multiply all numbers against all numbers in a specific sample, so it’s heavy.