Mark Williams-Cook

Yeah.

Yeah, it's, I think it's really, you can apply that leaky bucket analogy, which are probably going to regret saying off the cuff now on that podcast.

Now I've been cornered about it, but it's really just about how a lot of systems that are becoming reliant on large language model based technology.

And I say that specifically rather than just the general term AI.

is because the output of a lot of these systems is non-deterministic you know most of the time you'll get this answer but sometimes you get something a bit spicy and wild it creates systemic issues for chains of things that need to be correct over and over and

that leads in my opinion to all kinds of trouble with search engines because firstly the scale i think google said recently they did eight trillion searches last year like i know that's how many searches they're doing right

So even if you had a very small percentage of the results they were giving as wrong, that's still an awful lot of people getting incorrect or faulty in some way information.

And tied to this, and it's tied to how the user interface, I think, of a lot of the, whether you want to call them like chat assistants or AI search surfaces work, is there's a strong user feedback loop.

Even with traditional search, they watch very closely how people interact with their search result pages, what they're scrolling, what they're hovering on, what they click on when they come back.

And that works quite well when you have a kind of static page of results.

There's a lot of parity between searches and the user input is still messy and noisy.

but you've got some control over it.

I think the issue with providing things like AI overviews in Google search is that users, for a variety of reasons, maybe won't realize the information is incorrect, but it looks like it's correct and it looks enticing and they're satisfied with it.

And we start getting this impact of...

we're reinforcing incorrect information or bad information and that starts to grow and then someone else starts picking up on how that works.

I mean, we already see it in the industry in terms of there was a lot written about generative engine optimization.

which people basically were just guessing.

Because as soon as anything changes in SEO, there's an ultimate guide to it like two days later before anyone understands it.

But then, of course, all of the AI search services drink this up as the answer.

And then a new wave of people are like, how does this work?