This Ain't No Content

Being Human

This is prompted by another blog post, Thoughts on thinking, which surprised me. It's short, so you should read it rather than trust my summary. But basically, he thinks that LLMs write better than him and, because he uses writing to think, he feels his ability to think rigorously is atrophying. That is a scary thought.

Why am I surprised? Because he says he's being using LLMs to explore new thoughts and ideas for the last couple of years. It doesn't match how I've beeing using them. To me, the single biggest use of LLMs outside coding (and that is 95% on home projects) is helping out when I can't quite remember enough to help my son with his high school maths. It's great for that. Of course, you have to use it judiciously, as if it were a particularly kind and patient tutor, because the risk is that it will just do your homework for you and you learn nothing.

This is similar to the risks the fine blog post talks to, except that he's talking about his own thinking. I'll have to be honest and say it never occurred to me to use an LLM that way. But, in any case, I can see the danger by analogy.

There's also an analogy I remember from my own highschool days. I once did a mock Leaving Cert exam in maths and got an A-. The mock used real exam questions from previous years and I had studied them. In a phenomenon I have since seen described as "overfitting", I trained myself too specifically on those questions and not enough on the surrounding context. I got a C in the real exam a few months later.

I want to quote the last paragraph:

And I’m still stuck. But at least I’m here, writing this, and conveying my raw thoughts directly into your brain. And that means something, I think, even though an AI could probably have written this post far more quickly, eloquently, and concisely. It’s horrifying.

Yes, the AI would have written it far more quickly. I really doubt it would have written more eloquently. LLMs are good mimics, but they are not human. I'll admit, I haven't tried to get an LLM to write an essay for me, but I haven't been too impressed with samples that I've seen other people publish. It's much more obvious with AI artwork that some people use to illustrate their articles. It is superficially better quality than what they could have done before, yes, but not as good as a real artist would have produced. (I suspect that a real artist could produce AI work that would be indistinguishable from the non-AI version, but that's not what we're talking about.)

In a way, I'm not disagreeing with him at all. I do think that personal voices matter. It's why I have a blog and I think more people should. (I do see the irony of this every time I look at who reads my work -- it is almost all bots, many of them bots that feed LLMs.)