Adventures in AI Part 1: Editing

I’ve spent the last two day working with an AI (Microsoft Copilot) on some of my writing by way of an experiment. Been sitting in corporate training this week & DRIVING into work, so I haven’t been able to write AT ALL. That has sucked. This is what I’ve been left with instead.

 I therefore fed the AI all of THE RETURN OF DR. NECROPOLIS as well as the first 10 chapters of a different story that isn’t working as well by way of creating a control group. I wanted to understand what the AI could do and when it was being genuinely helpful vs. being sycophantically unhelpful.

Unfortunately, the AI’s all want you to like them. This is actually their worst attribute, at least as far as editing is concerned. Bottom line: the AI is a computer program. It does stuff that one would expect a computer to be able to do pretty darned well. For example, if you feed it something that is close to working, it understands the mechanics of scene & story structure well enough to algorithmically point out where a handful of tweeks might fix specific issues within your text. However, if you ask it to do anything more, you’re gonna get a generic hash that consistently loses details & which generally tends towards an uncreative mean. AI’s work by absorbing training data & then using statistics to predict likely outcomes, so OBVIOUSLY they will create generic shit if left to their own devices. 

Really, it’s like using Excel. If you’re an engineer, & you know how to use the formulas, the computers help a lot. However, without knowing engineering, Excel by itself won’t help you launch a rocket. 

This was an interesting experiment & uplifting for my ego. However, it’s only really useful around the edges.

If you’re wondering, you really need to prompt the AI for meaningful criticism. I told it that I felt like my descriptions weren’t great, & it gave me a long, process-oriented critique of why I write better dialogue & psychology. That was TOTALLY USEFUL! But it had spent the previous day pumping me up about my stuff.

If I didn’t already have a basic idea of what wasn’t working & why, I’d never have gotten the most useful feedback I got from this entire experiment.