AI is ruining science
For context, know that I used to work in academia, specifically researching the optimization of deep learning models (NOT LLMs). I won’t go into more detail so as not to dox myself, but I’ve worked at R1/Ivy institutions and am what would be considered an expert when it comes to AI.
Now, why am I writing this post?
LLMs are everywhere in academia, and it sucks. They’re much more widespread than you think. At the best facilities in the world, I have colleagues developing what I would argue is AI psychosis. They’re sending me snippets of syllabi that have been entirely written by AI, talking about how insightful it is. They’re designing experiments which they are not qualified to design and which are frankly poorly done. We’re getting peer reviews back that are obviously LLM slop. Hell, there are published papers in respected venues that you read and it’s clear as day they were written entirely by a chatbot.
Oh, and this is everywhere in the classrooms too, not only for undergrads but graduate students as well. You hear that? The people we are training to have PhDs are readying for their quals by being fed AI slop. The only people holding out are those who don’t know how to use the technology or those who have some kind of chip on their shoulder. These people are the exceptions and are going to be judged if their output is not as fast as their peers.
I don’t want this to be a doomer post, but I am worried. Science is built on top of itself. When I publish a finding, it’s predicated on the assumption that previous findings were well vetted. By using LLMs, which are designed to sound convincing, we are directly degrading that process. I hope I am wrong, but I think the reckoning for what we’re doing is going to be catastrophic.