3 ms·
It’s shocking to see so many famous intellectuals like Stallman and Chomsky get this so wrong. It seems like they have preconceptions so strongly held that no a
by _yb2s 4y ago
It’s shocking to see so many famous intellectuals like Stallman and Chomsky get this so wrong. It seems like they have preconceptions so strongly held that no amount of hard evidence could convince them otherwise.
GPT-4 deeply understands what it is talking about. I was able to pose it difficult classical physics problems that would be hard for a physics undergrad to solve and it could give correct answers consistently. These are things I made up that are nothing like existing examples. When asked to explain it’s work, or modify it for unusual scenarios, it could do so. I could also get it to invent new terms and concepts to describe the work, and then use those correctly. It even could reason out appropriate simplifying approximations when demands exceeded its calculating abilities. If gpt-4 isn’t really understanding what it is saying, or not intelligent than neither are any humans.
- DoItToMe81 4y agoBecause it's assembling data from an enormous dataset with correct answers. It's a language model, it is physically impossible for it to 'think' or 'feel'.
- _yb2s 4y agoThose answers aren't in the training set, nor is anything like it. It's assembling a correct and general model of physics from examples in the training set, and applying that understanding to be able to solve any arbitrary problem. Nobody knows what "think" or "feel" really mean, and I think this surprising phenomena is a clue to understanding that- it isn't what it seems. Most of our theories about what the human mind, language, and thinking are imply this should be impossible, yet it's clearly not.