Tag: language prediction algorithms

I Am a Language Model.

Posted in philosophyTagged , , , , , , , , , , Leave a comment
1–2 minutes

I was writing a video script, and a warning popped up. I first noticed this warning a couple days ago, but I didn’t think anything of it.

‘Strong resemblance to AI text.’

I am creating some videos to promote my latest book. I realise that it is academically dense, so I want to break down some of the subject matter.

As usual, my writing is on language and its insufficiencies.

But why was I being warned that my writing resembled AI? What does that even mean?

I reread the sentence:

What if the biggest trick language played on you is convincing you that the world is made of things?

I was looking for a hook to open the short clip. I altered it slightly:

What if the biggest trick language played on you is convincing you that the world is made of things, of objects?

This appeased the AI detector. I had to dilute the message by adding ‘of objects‘. As I write this, it reminds me of Margaret Atwood’s A Handmaid’s Tale and the naming of the women as property of their masters – OfFred, and so on. Of objects.

The AI police are annoying to say the least. Profiling: Minority Project. A 1984 thought crime.

I’m hopping down off the soapbox, down off my high horse, but I’m miffed by bollocks.

How does AI summarise it? Find out here:

Audio: NotebookLM summary podcast of this topic.

EDIT: Does it get worse?

When I checked into LinkedIn to share this post, I was distracted by another thread chatting about Emotivism. I’ll spare you the entire thread, but now Grammarly wants me to write in German. Was ist los?

On Predictive Text, Algebra, and the Ghost of Markov

Posted in language, maths, philosophy, technology, videoTagged , , , , , , , , , , , , , , , , , 1 Comment

Before I was a writer, before I was a management consultant, before I was an economist, and before I was a statistician, I was a student.

Video: Veritasium piece on Markov chains and more.

Back then, when dinosaurs roamed the chalkboards, I fell for a rather esoteric field: stochastic processes, specifically, Markov chains and Monte Carlo simulations. These weren’t just idle fascinations. They were elegant, probabilistic odes to chaos, dressed up in matrix notation. I’ll not bore you with my practical use of linear algebra.

So imagine my surprise (feigned, of course) when, decades later, I find myself confronted by the same concepts under a different guise—this time in the pocket-sized daemon we all carry: predictive text.

If you’ve not watched it yet, this excellent explainer by Veritasium demystifies how Markov chains can simulate plausible language. In essence, if you’ve ever marvelled at your phone guessing the next word in your sentence, you can thank a Russian mathematician and a few assumptions about memoryless transitions.

But here’s the rub. The predictive text often gets it hilariously wrong. Start typing “to be or not to—” and it offers you “schedule a meeting.” Close, but existentially off. This isn’t just clunky programming; it’s probabilistic dementia.

This leads me to a pet peeve: people who smugly proclaim they’ve “never used algebra” since high school. I hear this a lot. It’s the battle cry of the proudly innumerate. What they mean, of course, is they’ve never recognised algebra in the wild. They think if they’re not solving for x with a number 2 pencil, it doesn’t count. Meanwhile, their phone is doing a polynomial dance just to autocorrect their butchery of the English language.

It’s a classic case of not recognising the water in which we’re swimming. Algebra is everywhere. Markov chains are everywhere. And Monte Carlo simulations are probably calculating your credit risk as we speak. Just because the interface is clean and the maths is hidden behind a swipeable veneer doesn’t mean the complexity has vanished. It’s merely gone incognito.

As someone who has used maths across various fields – software development, data analysis, policy modelling – I can tell you that I use less of it than a physicist, but probably more than your average lifestyle coach. I say this not to flex but to point out that even minimal exposure to mathematical literacy grants one the ability to notice when the machines are quietly doing cartwheels behind the curtain.

So the next time your phone offers you a sentence completion that reads like it’s been dropped on its head, spare a thought for Markov. He’s doing his best, bless him. It’s just that probability doesn’t always align with meaning.

Or as the algorithms might say: “To be or not to – subscribe for updates.”