PSA: Neural networks aren't actually learning like humans do, stop saying they 'think'

I keep seeing posts on here about how AI models are 'thinking' or 'reasoning' their way through problems. That's not how it works. I work with LLMs daily and they're just statistical pattern matchers. Last month I tested a model on a simple logic puzzle that a 5th grader could solve, but it failed because the wording was slightly different from its training data. If it was really thinking it would have figured it out. The hype is getting out of hand when people start saying machines have consciousness. It matters because companies are making decisions based on this false idea. Has anyone else tried giving a model a basic reasoning test and watched it completely fall apart?

2 comments

2 Comments

the_rowan4d ago

Tried giving one a simple word problem and it guessed wrong three times in a row before I gave up.

clark.joel4d ago

Oh man, that reminds me. I once asked one of those customer service chatbots to help me find a specific pair of shoes on a website. I told it the brand, the color, and the size. It kept showing me the wrong model, like three or four different times. Finally I just gave up and did the search myself. It took me like 30 seconds to find them. I don't get why they make these things so hard when a simple search bar works fine. It's like they're designed to waste your time on purpose.