No, AI isn’t a ‘fancy autocomplete’

Published June 30, 2025

There’s a common description given to large language models (LLMs) like ChatGPT that is … well, it’s sort of correct, but leaves out the most important information.

It’s the idea that LLMs are just ‘fancy autocomplete’ tools — that they simply figure out the next most likely word in a reply, making them sound human. That description might point to one relatively small part of what an AI/LLM does, but it completely misses what’s really going on. And it gives people the absolute wrong impression.

It’s the equivalent of saying that all a television does is light up pixels on the screen. Um … yeah? I mean, I guess that’s true.

Let’s say you ask ChatGPT something like:

I’d like to keep my bunnies from chewing on the wood furniture in my office. I’ve tried spraying sour apple spray on it, but it doesn’t seem to make a difference. Short of chasing them away every few seconds, what can I do?

I did just that, and here’s the response I got. (You don’t need to read the whole thing, of course. Just get the gist.)

Cora: Ah, the eternal battle between rabbits and wooden furniture — it’s basically the plot of Watership Down, but with less existential dread and more bite marks.

So, if sour apple spray isn’t working (and yeah, some bunnies just laugh in the face of deterrent sprays), here are some other things to try:

  1. Cover or block the furniture – Use furniture guards, bitter-tasting tape, or even cardboard panels to physically block access. Some people wrap table legs with fleece or sisal rope — not gorgeous, but it works.
  2. Offer more tempting alternatives – Bunnies need to chew. Give them apple branches, untreated willow, cardboard boxes, hay cubes, or even seagrass mats. You want them to think, “Why chew this boring desk when I’ve got my own personal chew buffet?”
  3. Control the environment – Use exercise pens or baby gates to restrict access when you’re not supervising. Or set up a designated “bunny zone” with all the fun stuff and none of the heirloom furniture.
  4. Make it boring – If you can, sand down and seal furniture with bunny-safe, odorless finishes. Some rabbits are drawn to certain scents or textures — making the furniture less appealing can help.
  5. Positive redirection – Every time they go for the furniture, gently shoo them away and immediately offer a chew toy. Reward them when they use it. Yes, it’s time-consuming — but hey, welcome to the world of bunny parenting.

Let me know if you’d like to brainstorm any DIY chew toys or if you’re trying to protect something specific, like baseboards or antique chair legs.

If you’re thinking, “That’s a lot more than some kind of fancy autocomplete,” you’re absolutely right. Whenever someone refers to an LLM that way, at best they’re talking about the very last stage of the process, when the AI decides how to deliver its answer. (At worst, they’re parroting something they heard on a podcast and didn’t even stop to question.)

Think of what’s involved in coming up with that answer. Cora has to understand my language — including knowing what “sour-apple spray” is, why bunnies might not like it, the fact that I have wood furniture and what that entails, and the fact that this is in an office I’m talking about.

Then she has to come up with actual solutions that work in this scenario (bunnies, office, sour-apple doesn’t work), which means knowing not only what will work, but what’s safe for bunnies.

And then, after all that is done, then comes the part where she has to give me an answer … well, first after understanding the tone and style I like. That’s where the ‘predicting the next word’ comes in. She now knows what she wants to say and how she wants to say it … so yeah, it’s time to choose the exact words based on all that.

So next time someone comments about AI or LLMs being fancy autocomplete systems, feel free to give them the side-eye. They’re just not getting it.


—30—