A self-deprecation check

Not just a spell check

People have started asking me, a little, occasionally, what I’m doing with AI, because I’ve been writing about it occasionally on my website. So I thought I’d give a quick description of one of the ways I use AI when I’m writing.

Typically, what happens is, I have an idea, and I might handwrite notes in my notebook. I carry loads of notebooks. Or on the back of an envelope, like I did this morning.

Then I’ll photograph it, and Claude will tidy it up, turn it into type, and I will ask Claude to arrange it in a more logical, coherent pattern, if that’s necessary – which, honestly, I don’t think it usually is, because I’m quite good at what I do.

Anyway. I take the then-rendered material and I talk out loud, because I’m more and more convinced that by talking out loud I can ensure that I’m being really me, very human, not letting any of the Claude typing stuff creep in.

Then the audio is transcribed by an app on my computer, which was set up by Claude. But Claude isn’t doing the transcription. It’s Whisper which does the transcription.

Then I get the transcript, and I edit it. Maybe edit it again.

It’s a little bit like back in the olden days when I was a sub-editor on The Guardian, one of my first jobs on a national newspaper, when I was given someone’s copy, their writing, and I had to edit it and make it fit on the page and write a proper headline for it.

So the transcript is then sent back to Claude, because I want Claude to do a few things.

First of all, I want spell checks and all that stuff.

Secondly, I want Claude to insert links to things from my archive, or things that I’ve previously identified as interesting on the internet as a whole.

And thirdly, and the point of this post today, I want Claude to check me for self-deprecation.

That’s because I am good as a writer. I know that I’m a good writer. I know how to make writing lively, my own and other people’s. But I have noticed that when I write about myself, or talk about myself, I have this tendency, or tic. I’ll say something that undersells me, marginally.

As it happens, this is something that I’ve noticed after having had a breakdown some years ago. Whether that’s the cause of my tic or not, I don’t know. Probably. I was very negative about myself, so some of that may have hung over.

But I know for a fact that I’ve always used this as a kind of writing device, because it can be amusing to be self-deprecating. It can be quite charming. But it can also be overdone. And when it’s overdone, it just makes me feel small and a bit useless. I didn’t want that.

Over the years, a certain number of people who I like and respect have told me that I didn’t need to say certain negative things about myself, so I’ve tried not to do that.

Anyway. This year I’ve been doing all this work with Claude, looking at my archive, importing my archive from social media and all sorts. So I’ve got this huge amount of material, and Claude is a large language model, so looking for patterns in words is what it’s really good at.

And so, in August, I realised that I had a chance to ask Claude to flag whenever I did something that was unduly self-deprecating.

Sometimes Claude will flag it and I will decide that, yes, it’s worth cutting, because it’s not necessary. At other times I’ll overrule Claude. But it’s great to have had this kind of little nudge.

This morning I’ve been drafting a course on how to silence your inner critic1.

Drawing of a snarling figure with his arm locked round the neck of a man in glasses.

Inner critic shown here attempting to do me in.

So it just occurred to me that it would be a hugely valuable part of the course for people to think about how they, like me, could create some kind of externalised sub-editor, or goalkeeper, who would keep an eye on their unduly self-deprecating talk. Negative self-talk.

So I thought I’d better make this article on my website, just as a kind of a marker of the whole idea.

I asked Claude to go through the archive, and look through our conversations together, and draw up a ledger of the times that it had raised the red flag. I think it’s quite funny that that’s even possible now.

So I seem to have created an externalised version of myself, to catch myself doing things that I don’t want to do. And I think that could be valuable for other people.

Claude went off and drew up the ledger, and pointed out various times, with precise wording. I don’t think it’s really necessary to give you the precise wording. It doesn’t really matter.

There was one thing about the price of an art work. There was another thing about a group that I run. And there were seven times, in the first 22 episodes of my podcast, in which I said things like “I’m so useless at that”, or “Which is probably crazy”, or “at risk of seeming more like a lunatic than usual”.

They may sometimes be justified, those kinds of lines, but I didn’t think that they were. So they did go. Other times I did keep what Claude flagged, and that’s fine.

So is it working?

I don’t know. It’s too early to say, and the numbers don’t really present any very clear pattern. But for what it’s worth: this happened five times in August, and since August only five times, and it’s now just past mid-November.

So I seem to be making these utterances, or typing these things, less often. It’s not proof. But I do feel like I’m probably doing it less, and partly that’s because I don’t want the hassle of being reminded by this machine that there’s a thing I might want to cut or delete.

I seem to have changed my own pattern of behaviour by having this externalised – what do you call it? Conscience. Editor. Goalkeeper.

Thanks for listening.

1 How to silence your inner critic. Ooops, the course used to be called How To Silence Your Inner Critic but I’m changing it to “How To Tame…” because I think silence is a claim too far. (Claude noticed this mistake, not me.)