It makes bad stuff better.

It makes poor writers better. So too for unskilled graphic designers or amateur financial analysts.

But also spam, scams, and cyber threats. It used to be easy to see a phishing email for what it was, but the Nigerian princes are gone.

At the same time, it also makes the great stuff worse.

Great writing becomes banal and repetitive. AI art looks like… AI art (or the boring stuff you see in hotels).

It brings the good stuff down while bringing the bad stuff up. In short, it averages it out.

And that makes sense because LLMs imbibe everything that exists—the great, the good, the average, and the bad.

And in most subjects, the bad always outnumbers the great (or even the good). This lowers the average.

The key is to know what to use it for.

If you’re bad at something, AI will make you better at it. So use it.

But if you’re good (or especially if you’re great) at something… Think twice before handing it off to AI.

I ran this post through Claude and, based on its knowledge of my style, asked it for feedback. Claude told me the post had great bones, but it also told me to remove certain items and phrasing… Precisely the same things that make my writing style what it is.

It told me to remove a tiny explanatory sentence as well (do you know which one?). And I refused to do it, because there are plenty of people who still don’t know how this stuff works. That knowledge is central to the premise of this post.

It also recommended that I end the post with something like, “Average is the enemy of excellence.” How many times have you heard that before? It felt too much like a motivational poster for me—I can already see the kitten attempting pullups with that as the headline.

So, I think I should actually end this post by saying, “Claude, your suggestions were appreciated, but wrong.”¹

And that is precisely my point.

---

1. I did make a couple of tiny grammatical changes it suggested, which did actually improve the post. And sometimes, Claude’s recommendations actually do improve my writing.
   
   And, somewhat ironically, Claude did improve this post by giving me feedback, even though the feedback turned out to be flawed. It proved my point in real time. So I must give credit when it’s due.
   
   But more often than not, Claude’s suggestions make my writing worse because it no longer sounds like me. ↩︎