We do know in general how LLMs work. Now, it's of course not always possible to say why a specific output is generated given an input, but we do know HOW it doe

We do know in general how LLMs work. Now, it's of course not always possible to say why a specific output is generated given an input, but we do know HOW it does it. To translate it to your analogy with dogs: We do know how the anatomy of dogs work, but we do not know why they sometimes fetch the stick and sometimes not.