It’s one thing to think it’s obvious, but quite another to prove it. I think this is the true value of this kind of work, is that it’s helping to decipher what

It’s one thing to think it’s obvious, but quite another to prove it. I think this is the true value of this kind of work, is that it’s helping to decipher what these models are actually doing. Far too often we hear “NNs / LLMs are black boxes” as if that’s the end of the conversation.