Model Numbers Don’t Matter (Much): Your AI Is Only as Smart as You Are

Think before you... AI.

Every time a new LLM drops, the internet acts like we’ve all been handed the keys to a Ferrari. ChatGPT 5 Thinking! Gemini Ultra! Claude 4.1! Surely this means our work will be faster, better, and smarter—no extra effort required.

Here’s the problem: a model upgrade doesn’t make you better at using it. And unless you have the skills to guide it, the output from “the most powerful model in the world” might be worse than what you could get from last year’s version.

Don’t believe me? 

Well, try this. Your job is to get the LLM to write a personal email in your style. The recipient is somebody you know well, so that corporate boilerplate speak is out the window. 

How would you do it?

“I’d prompt it,” you might say. If I pushed you to be a little more specific, there’s a chance you might say something like: I’d ask it to write in my style, describing how I write, the kinds of words I use, and the overall tone I’m going for.

Sounds reasonable, right?

Wrong. You’re going to get an email that sounds like a generic idea of you. “But that’s just AI,” one might object. 

The better approach is something called context layering, which can be summed up as the text, image or video examples you provide. Context layering helps LLM function far more effectively at a task. 

With the email example above, the context should be pretty obvious: you don’t tell ChatGPT how you write, you show it by uploading several of your emails. You let it infer the patterns and cadence that it can then spin into something that sounds a lot more like you—at least a lot more than telling it what you think your writing sounds like.  

AI Technique is More Important than AI Model

This post isn’t really about context layering. That is only one of several powerful techniques that many simply aren’t using. Not using them vs. using them can be far greater than the difference between models years apart. 

Now let’s imagine we turn the above into a contest. In Corner 1 is somebody who knows how to do context layering, not to mention several other clever techniques to refine writing. In Corner 2, we have the average ChatGPT user. 

The twist: the contestant in Corner 1 is stuck using ChatGPT 4. Yep, plain old four, no lower case letters or decimal points appended. Whereas Corner 2 has all the power of ChatGPT 5 Thinking, the most powerful model on the planet. 

Whom would you bet on?

Before you answer that, I’ll share a real-world story. 

We Got Ambushed by ChatGPT 5!

My team and I have been tasked to create test prep questions for a client. Interestingly, ChatGPT happened to debut in the middle of this project. We’d been using 4.5 and o3 to create batches of questions. I wondered: how much better would ChatGPT 5 do? Would it be the ultimate thinking machine? 

I uploaded some of the questions we’d created into a ChatGPT5 Thinking thread. Was the power of all things text-based suddenly at my fingertips? 

Well, ChatGPT 5 did not think very highly of the questions, even calling some of the more difficult ones far too easy. This meant that questions we created with our many years of experience in test prep were off the mark, as were what we like to think are our most innovative GenAI techniques. 

Someone without our many years of experience might just have given in to ChatGPT 5 Thinking’s decree. But I smelled something off in the reasoning it provided. I then asked it to create what it thought was more representative of the test. It created a question fit for a Math Olympiad—not for a test that tests high school math concepts. 

But how? How could ChatGPT 5 do something so seemingly amateur?

Well it wasn’t necessarily ChatGPT 5’s fault. In my rush to tap into the seemingly omniscient oracle that OpenAI released late last week, I forgot one of our baseline teaching’s: context layering. Or show don’t tell. 

I didn’t first feed it official questions from the test to calibrate its knowledge. It went off what the web says about the test—info that can be all over the place. And turn what should be a question a smart high school senior should get to one that would confound a Ph.D math department. 

The Takeaway

Why does this all matter? First off, many of us seeing how powerful the LLMs are getting—and all the hype and anticipation behind ChatGPT 5’s release certainly helps fuel this—will be tempted to think they have to do even less because the models can do even more. 

But if you take the human know-how out of the equation, LLMs can lead you dangerously astray, giving you content that is inaccurate, be it a client deliverable or a personal email. And it is that human know-how that can make a far bigger difference than the delta between multiple generation gaps in LLMs. 

So give me ChatGPT 4 circa April 2023 any day over ChatGPT 5 or Gemini’s next big splash— as long as I’m using the proper techniques based on years of learning how to make these LLMs sing. Of course, I’m not forced to make that choice. None of us are. We can all use ChatGPT 5. 

Just remember: you can’t outsource skill to the model. Learn the playbook first, then let the model—whichever number it is—do the heavy lifting.