Brutkey

Martin Seeger
@masek@infosec.exchange

How do I use AI and handle AI-generated content?


I regularly use AI to discuss the texts I’m writing. For example, I want to see whether the AI understands what I was trying to say and how I played with certain ideas. If it doesn’t get it, maybe I should adapt my phrasing.

Sadly, the AI is often better at recognizing allusions than the average reader.

Sometimes I know an exact quote, but only in a different language from the one I’m writing in. That’s something I can verify with AI far more easily than with a search engine.

I don’t always have the right people around to discuss an idea with. An AI is no substitute for a competent expert on the topic, but it’s still far more useful than just letting the idea bounce around in my head for hours. It’s too crowded in there anyway.

My biggest issue when working with AI is that I’d prefer a grumpy professor over a cheerleader. AI feedback tends to be far too positive, and it doesn’t clearly separate important criticism from minor stylistic notes.

In my experience, AI works best on small text segments and when given plenty of context. Asking it to write long pieces is just asking for it to go off the rails. So I try to discuss it in small pieces.

But most important of all:
I take full responsibility for everything that comes out at the end. This goes far beyond avoiding hallucinations, it has to say exactly what I intended to say.

That’s also what I expect from others who use AI. I don’t mind AI artifacts like the “rule of three.”

I’d rather see a mediocre AI-generated picture at the top of a post than an advertisement with 397 tracking cookies attached. But be aware: it still reduces the quality of your statement.

What I can’t stand is text being inflated by AI from a half-statement just to fill a page. And if you use AI, you’re responsible for keeping it on the rails. If the AI spews nonsense under your name, I’ll treat it as if you had written it yourself.

So, in summary: I judge any text primarily by its intrinsic value. Does it provide new information, and is it dense enough? Do I like the style? Is it consistent and coherent?

If yes, I don’t care whether you used a pen or ChatGPT.

If not, any use of an AI was simply a waste of resources.