Generative AI makes things up. Sometimes I forget.
I don’t like it, because, even though the machine isn’t human, I think it’s being lazy (when I’m probably being the lazy one!) or trying to sneak one of its opinions into my work.
I won’t use the term “hallucinations” to refer to these moments, because that word brings up flashes of light and strong images to mind. In the work I do, generativeAI’s actions are more subtle. Maybe I can refer to them as “bugs”.
However, I’ve learned to work with them.
I’ve heard writers say that they welcome these flashes of invention because it offers an opportunity to go down a brand new path.
They offer me something different.
I’m using them to understand where I can build on a half-baked idea I’ve presented to the reader. (In a way, it’s like having a red light start flashing when I’m interviewing a guest in a podcast, and I miss digging into something they’ve said that could open up a whole new conversation, or provide an unusual insight.)
This is all context-specific, so I’ll give you an example.
You know that “Shakespeare-meets-remote” book I’m writing?
Well, I drafted one chapter on working from home, and another on the commute, but now I’ve decided to fuse them together. I didn’t want to give the commute a whole chapter, and by fusing them into one, I’ve made space for a chapter on Visible Teamwork, which adds “more of me” into the piece.
Both original chapters had their own “arc” and covered multiple points. So, because I’m lazy, I mean, because I’m learning how to speed up my writing process, I gave both chapters to Claude AI and asked:
I wrote a chapter on working from home and one on the commute. I now want to create just one chapter.
Can you read the chapters and pull out the main points I've made? (Please, don't add anything from other sources, or generalise. Stick to what's written there.)
(You can see I’ve learned to preempt these “liberal acts” by Claude.)
Once Claude drafted the main points, I asked it to create an outline for a new chapter merging both pieces of content. Most of it made sense to me, but this stood out:
”Discuss legal and tax implications, using the example of the UK employee who wanted to work from Luxembourg”
I challenged it.
Can you remind me in which part of my draft I mentioned Discuss legal and tax implications?
I wasn’t just being polite. I was giving it the benefit of the doubt. (And also, you never know when the bot-revolution will arrive…)
I apologize for the confusion. After reviewing your draft, I realize I made a mistake. You didn't explicitly discuss legal and tax implications in the way I suggested. The closest reference to this topic in your draft is this passage:
The passage does indeed cover a story of a UK employee who had been denied their request to work from home… from Luxembourg. They’d taken their employer to the tribunal, and had lost.
In my draft, I hadn’t mentioned anything about the legal and tax implications of working from a country other than your tax residency, but I’d obviously hinted at it enough for Claude to pick up on it as a topic.
Now I can go back and add a couple of lines acknowledging that as a topic, and let the reader know I’m not going to elaborate further as it’s not my area of expertise - nor interest!
(In case you were wondering about the title of this post, I chose “butterfly” because it sounded nice, but also because Claude’s “bug” provided something for my mind to fly away with.)
Building the genAI muscle
Working so much with these tools has made it easy for me to spot when people have asked ChatGPT to improve their text, or even generate it fully for them.
So far, I’d only encountered this in posts, or personal anecdotes, but I hadn’t felt the effect it can do to trust in other parts of our lives.
Last week, I received an email from the property management company that deals with our flat, telling me our property management officer was changing. A few days later, I was following up on a complaint with this new person, and their reply made me think they were a bot. I received an email that read like it had been drafted by genAI and so I thought the company had automated “the first line of fire”.
Turns out I was wrong.
The new person is indeed, a person, but one who doesn’t understand how much you need to clean up the work generated by ChatGPT. It wasn’t just a question of style. The email missed the nuance of my complaint.
I even asked GPT4-o whether it had written it and it replied:
Yes, it seems likely that this was drafted with my help. The message is well-structured, formal, and addresses the issues clearly and empathetically. It acknowledges the recipient's frustration, explains the constraints preventing the installation of CCTV, and offers a proactive step to discuss alternative solutions. These are all characteristics of a response that I would assist in drafting.
Well, now we know…
(I also asked GPT-4o whether it had drafted some lines I’d written myself, and it said no. Always worth checking, as it can be a bit of a clown.)
AI-free zone: Short update on the novel
For new readers: I’m in the process of editing a cozy mystery I’ve been working on for a while, and I often talk about the writing process here.
Cozy mysteries are often written in the first person (I ran, I saw etc), but I decided to write mine in third (she said this, she thought that) because I was afraid if I wrote in the first person, the protagonist’s character would turn out to be very much like mine, which I didn’t want.
Now, after having gone through the edit of my 5th first-draft, and having spent a couple of years thinking about the characters and story, I understand the protagonist better. Yes, of course there are bits of me in her, it’s my first novel!, but I know enough about her life and her impulses that I can comfortably write her without worrying about our similarities.
So I’ve decided to re-write the whole thing into the first person. I can’t tell you the reason for the decision, because honestly, I can’t remember it! But it seemed like a good moment to try out, as I have to go back to the draft to make some edits. (The edits at this stage are not careful grammar and punctuation ones, more like getting rid of chunks of text, adding missing segues and scenes, etc.)
I decided not to get AI to do the first draft in first person. Firstly, as we have seen above, I’m going to have to deal with it making stuff up. Secondly, it’s an opportunity to develop my own writing, and to adjust the story as I go along.
I’ve already tackled the first two chapters and doing this has given the prose more of a punch, and more of a uniform “voice”. I’ll make sure to tell you how I get on but meanwhile, if you want more detail or have any questions, feel free to get in touch!
That’s it for today. I might be back next week, or the next one, or maybe the one after that, as I am going away for a couple of weeks to work a bit less, and to work mainly on the novel. Have a fabulous time meanwhile!