LLMs Are Not Magic, They're Mirrors
Everyone asks whether large language models understand. The more useful question is what they reflect back at us.
Everyone keeps asking whether large language models understand. I think that is the wrong question.
The more useful question is what they reflect back at us, and what that reflection tells us about language, tools, and the way we build software.
The useful illusion
A model can produce a remarkably coherent answer without possessing the kind of internal experience we normally associate with understanding. That does not make it useless. It makes it interesting.
The model is often less like a mind and more like a very strange mirror.
The engineering challenge is to learn where the reflection is reliable and where it is distorted.
Comments (2)