AI can't do ME

There’s that scene in ‘The Trip’ where Rob Brydon and Steve Coogan argue about who can do the best Michael Caine impersonation.

Coogan knows his is more accurate and he’s desperate to show it off.

But Brydon refuses to let him interrupt until he’s finished his own routine showing how Michael Caine’s voice has changed over the years. “Let me finish!”

Coogan finally takes over, critiquing Brydon’s impersonation of Caine in the voice of Caine. “It’s not quite nasal enough the way you’re doing it.”

And arguably his impersonation is more accurate, but it’s not as funny.

It’s like ChatGPT and Claude.

I switch endlessly between the two, with a dollop of Gemini thrown in when I revert to Google.

I seem to be settling on ChatGPT for home and Claude for work.

ChatGPT has helped me reinforce a garden bench, fix a loose skirting board and put air vents in a cupboard. I’m very proud.

Then Claude is my partner at pretty much every stage of a brand strategy project. From the proposal to the information gathering to meeting summaries to discussion guides to research analysis to story building to the strategy itself.

It can also hold this mountain of evidence in one place for immediate access, which is invaluable.

It’s even got me into using transcripts, certainly for individual interviews. That way you can get immediate feedback straight after the event and then a more detailed perspective as you go through it yourself.

But where it doesn’t work is with the insights and ideas. If I look back at the end of the show, they all come from me. I think I need to think it myself to get excited about it, the tingle down the spine.

It’s like this blog. Just over a year ago I started using Claude to help on posts. For some reason the first ‘joint’ post came out kind of like a poem. If I remember right, it started with a title idea – ‘The Long is made of many Shorts’.

I was intrigued so I continued. The idea always came from me but I started giving Claude a rough plan and intro and getting it to help from there on.

We spent time on the plan. Beginning-middle-end, the three-act play. I then drafted and we edited together. I almost couldn’t wait to upload a new version to get the feedback. I can feel the urge now.

But slowly I realised the problem. The posts were starting to follow a formula, a recurring pattern. I mean, I do understand how AI works but it’s still easy to lose your way.

The point is I want my posts to sound kind of spontaneous, like I’m actually thinking something through in real time, which I generally am and I hope you might be doing too. Claude has never really got that even with the most immaculate of voice prompts.

So now I’ve got it. AI is an impressionist. Claude and ChatGPT are both really good and they’ll probably soon be running the world and curing all our ills.

But whatever it is that makes me me, they can’t quite do.

Like clumsy sentences like that.