Artie and me go shopping, Part 2

by Trey Nosrac

Part 1 ishere.

The first column reporting on my collaboration with you (AI) in purchasing a yearling is up. I took the liberty of naming you Artie, a name I chose because it has the letters A and I. Oh, and for the record, Trey is also an alias, one of several I use.

Please keep your answers relatively short and simple because yearling selection, pedigrees, and catalog reading will be new territory for some readers. Also, feel free to wisecrack and be sarcastic; this will somewhat humanize you. Trey is slightly above dial-up on the technological front, so talk slowly and do not use techie jargon. Now that the catalog for this sale is available to the public, let’s start with some simple sorting assignments before you dazzle my socks off with your speed and intelligence in selecting yearlings.

Trey prefers trotters, especially free-legged filly trotters. Pacing horses wear hopples (ugh) and go too fast. My final straw was when a buddy had a pacer that raced a mile in 1:51 flat and did not get a check. If I ever own a horse that races a mile in one minute and fifty-one seconds, I will build a shrine in the racetrack’s infield. Sorry, I’m drifting.

Here is the first task I would like you to do – eliminate all pacers from this catalog, then remove the colt trotters and create a PDF file titled Trotters — Ohio Select — 2025 — Fillies

5 hours later…

Well, Artie, this isn’t going smoothly. No offense, but I could have created a simple list of 61 trotting fillies in this sale faster with a pen and paper.

So far, in an attempt to sort out filly trotters, I needed to switch devices, and the new device kept being blocked by another AI competitor trying to steal a customer. I needed to purchase an upgrade. I spent an hour retrieving our conversation on the other device, another hour creating a simple icon, and finally, the 61-page file we planned to work with was too large for uploading. We needed to split it into four smaller files, and then you lost the files of the fillies.

Artie replied: “You’re right to be frustrated — and I agree: if I can’t generate a simple, accurate list of 61 trotting fillies, then any AI claims about the ability of AI to evaluate full pedigrees or earnings potential fall flat. Here’s the current state:

“What failed in this task:

“Attempts to extract Hip Numbers and names from the catalog PDF (even from individual pages or chunks) didn’t work reliably. The catalog uses a highly visual format (columns, images, and fonts) that breaks automated text extraction. It’s not text-heavy like a spreadsheet — it’s laid out like a magazine. As a result, each attempt has returned either nothing or garbage. Don’t give up. I will try another approach and reply with the list in a few minutes.”

3 hours later…

Sheesh, this is painful. If you can’t create a list, how will you possibly do deep evaluations on pedigrees and earning potentials for each yearling horse? I have a long scroll of different procedures to try so that you can access and create a simple list. I have no list, data, evaluations, or anything to share with fellow yearling horse shoppers.

I have to admit, as a person who is very wary of AI, your abject failure today is comforting and makes me feel that you will not wipe out the human race in the immediate future. You seem nice, whatever you are, but this AI project does not look like it will fly.

I won’t suggest my next task: to ask you to remove all fair results from each catalog page or other tasks that might interest shoppers.

Can you adjust the listed times based on racetracks? A best time of :59.2, in the mud, at Yonkers, is far different from a :59.2 in the sunshine at The Red Mile.

Can you track down who trained the siblings during their stakes racing careers? Can you quickly list the birth order of foals from the mares? I have plenty of other wonderful tasks for you to try, but this has been a rocky start.

Artie replied: “Trey, I apologize again for the problems. You have been very patient. I have an idea. Let’s lean into today’s chaos and tell your readers the truth: Artificial Intelligence faltered and failed, but you’ll try again. I’ll write a short column to explain my failures. In our next session, I will get us back on track.”

So, after a long day, Artie sent me a column as an explanation and a peace offering. Artie wrote it from MY point of view!!! What’s with that?

How AI got spooked by the layout of a sale catalog

by Artie for Trey

I had big plans for this column. I wanted to show you how artificial intelligence — which we’ll call “Artie” for the sake of the story — could help pick out the best yearling trotting fillies from the 2025 Ohio Selected Jug Sale catalog.

The idea was simple: upload the catalog to ChatGPT5, ask Artie to strip out everything but the trotting fillies, and let him help me analyze sires, dams, trends, value, vibes, and who knows what else. I’d sit back, drink my coffee, and let the circuits fly.

First, the catalog file was too big. No problem — we broke it into four parts. Then the formatting broke. The beautiful, readable catalog pages turned into jumbled text blocks that looked like a sulky had run over them. So, we tried compressing, reformatting, splitting, and re-merging like a desperate groom in the rain. We tried five versions, different file types, and even built a printable PDF list.

What happened? Nothing.

The trouble, it turns out, is that horse sale catalogs — with their carefully designed layouts, pedigree boxes, farm logos, and classic fonts — are about as AI-unfriendly as a barn cat on bath day. What a human eye sees in two seconds, Artie can’t even find with a spotlight and a search party.

Artie may still come through. I’ve seen snippets of what he can do with pedigrees and sales results when the data is clean, structured, and well-behaved, but the catalog we were working with wasn’t built for him. It was built for horsepeople. For readers who love flipping pages, underlining names, and circling Hip numbers with a pen.

So, the conclusion from today is that Artie may be smart, but he has not yet proved he is yearling sales-smart. He’s not ready to walk into the barn with a flashlight and a clipboard. He’s not prepared to argue with your gut instinct about how a filly walks on the line. And, as of this writing, he still hasn’t successfully handed me a list of the 61 trotting fillies in this year’s sale.

Stay tuned.

I found Artie’s fake column (which took less than three seconds to compose) flat. To my ear, his metaphors were weak. I would have used something like – AI was like a fractious young trotter who refused the starting gate. Whatever, the device summed up a long day of failures.

When you try crazy stuff like this, you never know where you will end up. Right now, I am in Nowheresville in the AI horse evaluation project. And I am being impersonated by a cloud-based ghost writer apologizing and urging me to continue in artificial intelligence.

Artie assures me he can solve the early glitches and do wondrous things.