Fyreside Newsletter - 7/21/25

Hey Reader,

Last week, I said I was going to use this week’s newsletter to say mean things about AI. I plan to stick to that promise, so consider this your warning. If you’re a big AI fan, maybe skip this issue. Or don’t. I’m not your dad. If you're sticking around, strap in. It's a long one.

AI Sucks and I Judge You for Using it.

I’m talking specifically about generative AI here, so your ChatGPTs, your Copilots, etc. I’m also not saying that other types of AI are better than generative AI, that’s just not what this newsletter is about. This newsletter is about marketing, a field which, as of late, has become riddled with AI. And it all sucks. All of it. Every single one.

I acknowledge that there are some halfway decent use cases for AI across a broad spectrum of industries, but again, that’s not what we’re doing here today. If your eyes are bulging out of your head because, “WeLl WhAt AbOuT tHiS oNe ThInG?!,” go write your own newsletter.

Why AI Sucks.

I’m going to tell you my exact opinion on generative AI and why I think it’s a pathetic excuse for “innovation,” but I must first acknowledge a few reasons why AI sucks that have nothing to do with my opinion.

All of the above are reasons why AI sucks, but they’re not my opinion. In fact, I think AI sucks for a completely different reason: it strives to be average.

Large Language Models.

Do you know what ChatGPT really is? Or any of the colloquial generative AI tools? Well, let me tell you, they’re Large Language Models. So what’s a Large Language Model? It’s a large amount of language organized to make a model. I’m not trying to be a smartass here, that’s really all it is.

Let me put that differently, an LLM is a computer program that has “read” massive amounts of content (books, webpages, articles, etc.) in an effort to identify and replicate the patterns within that content. When you hear someone say they’re “training an AI,” what they’re really doing is scanning millions, if not billions, of pages into a machine that’s organizing it into a spreadsheet. Then that machine uses its internal spreadsheet to produce an “algorithm” that can reproduce any patterns it’s recognized.

Generative AI does not “think,” nor does it “create” or “produce” anything. The only thing an LLM is capable of is replicating a pattern. If you haven’t assembled a LEGO, sewn a quilt or purchased furniture from Ikea in the last 5 years, it may surprise you to learn that patterns can be incredibly complicated. The driving force behind generative AI (and why they get away with calling it “generative”), is nothing more than an exceptionally complex pattern. It would take a human being approximately 36 hours of nonstop reading to finish every novel that Stephen King has written, but ChatGPT could do it in minutes.

Here’s the thing, a human being probably doesn’t need to finish all 65 Stephen King novels to identify the patterns within them. An LLM does need to finish all 65 novels and it still won’t identify patterns as accurately as a human being.

No Child Left Behind.

If you didn’t have the misfortune of knowing me between the ages of 7-18 you may not know this, but I was really good at school. Please take note of how I structured that sentence. I was really good at school. That’s not me saying I was successful, happy, healthy, or anything else, just that I was good at getting good grades. Why, you ask? Because I went to a Blue Ribbon school!

If you’re not familiar with the disaster that was No Child Left Behind, the take away you need is that it birthed the “Teach to the Test” style of curricula. Students no longer needed to “learn,” they simply needed to pass the standardized test. Well, thanks to a decent memory and a penchant for listening, I was really good at passing standardized tests. Unfortunately, I have since learned that “passing standardized tests” does not “pay the bills,” but that’s a story for a different newsletter.

The point is, ChatGPT and any other LLM is just the personification of a “Teach to the Test” curriculum. As explained above, an LLM identifies patterns in whatever training materials it’s fed. If you ask it to replicate a pattern within its training materials, it should be able to, but heaven forbid you ask it a question that wasn’t on the test. And maybe that’s enough for you! Maybe you felt really good about No Child Left Behind! Maybe you thought the FEMA response to Hurricane Katrina was effective and well orchestrated! Who’s to say?

Why I Judge You for Using AI.

I want to go back to why I think AI sucks: it strives to be average. That’s what patterns are: the average, the thing that appears the most, what everyone else does. The absolute best thing an LLM can ever “create” is an amalgam of the most common ideas. It can’t be creative, it can’t be new, it can’t do anything but replicate patterns. I won’t speak for you, but I don’t settle for average.

In the world of marketing, the last thing that anyone should be doing is more of what everyone else is doing. The perpetual rat race to be the most average thing disgusts me. I hold myself, and every single thing I do at Bahnfyre, to a higher standard. You should too.

-MC

p.s.

Have a friend that HATES AI? Send them this newsletter!
Have a friend that LOVES AI? Send them this newsletter!
If someone forwarded you this email and you liked it, subscribe to this newsletter!