OpenAI's new Academy wants to teach you how to talk to robots (and I've seen this before)

The teacher guide is particularly interesting because it includes, and I'm quoting their description here, "the efficacy of AI detectors, and bias." Which means OpenAI is now in the business of explaining to educators why the tools meant to catch AI-generated work don't really work, and also why their own system might be biased. That's a lot of caveats for a product that's supposed to be transformative!

Here's what the Academy materials actually tell us:

• Users aren't getting good results without training (otherwise why build this?)
• The system requires specific communication patterns to work well
• Different professional contexts need different approaches, suggesting the "general intelligence" isn't quite general enough
• OpenAI is worried about enterprise adoption stalling
• The company knows teachers are struggling with how to handle this in classrooms

That last point matters more than the others, I think. When you're building guides specifically for educators about bias and detector limitations, you're basically acknowledging that your technology has created problems it can't solve.

Now, some of you kids (and I use that term with affection for anyone who started their career after smartphones existed) might argue this is just good customer education. Every enterprise software company has training programs! Salesforce has Trailhead! Microsoft has certifications!

Fair point. But those programs exist because the software is genuinely complex, with thousands of features and configuration options built up over decades. ChatGPT is, at its core, a text box. You type words, it types words back. The fact that using it effectively requires a curriculum suggests something fundamental about the gap between what these systems can do and what users expect them to do.

I've seen this movie before, actually. The early days of search engines had the same dynamic. Remember when people thought you needed to learn Boolean operators to use AltaVista? Remember "power user" guides for Yahoo? Then Google came along and you just typed what you wanted and it mostly worked. The complexity got hidden, the interface got simple, and the training programs went away.

We're not there yet with AI. The Academy exists because the technology still requires users to meet it more than halfway.

The manager and customer success guides are revealing in a different way. OpenAI is explicitly targeting enterprise workflows now, teaching people how to use ChatGPT for feedback conversations, account management, reducing churn. This is the company betting that AI assistants will become embedded in white-collar work the way spreadsheets did in the 80s.

Maybe they're right. But spreadsheets didn't need a guide explaining their limitations and biases. VisiCalc didn't ship with a warning about when not to trust its calculations. The fact that OpenAI feels compelled to be upfront about these issues, at least in the educational materials, suggests they know the hype has outpaced the reality.

And look, I'm not saying the Academy is bad. It's probably useful! If you're going to use these tools, you might as well learn to use them well. The writing guide apparently covers "clear structure, tone, and intent," which sounds like basic composition advice that happens to apply to talking to a language model. The personalization guide teaches you about memory and custom instructions, features that most casual users probably don't know exist.

But useful and telling aren't mutually exclusive. The existence of this curriculum, its scope, its professional specificity, tells us that we're still in the early awkward phase of human-AI interaction. The phase where the humans have to adapt to the machines because the machines can't quite adapt to us.

I covered the early days of autonomous vehicles (still covering them, actually, because they're still in the early days). The pattern is similar. Lots of hype about how the technology would just work, followed by years of edge cases and training data problems and the slow realization that "almost there" in AI terms can mean decades.

Will prompt engineering become a permanent skill like coding, or will it fade away as the systems get better at understanding intent? Remains unclear. OpenAI is betting on the former by building formal education around it. But I've watched enough "essential" tech skills become obsolete to know that's not guaranteed.

The Academy launches at an interesting moment. Enterprise AI adoption is real but slower than the hype suggested. Teachers are genuinely struggling with how to handle AI in education. The initial wow factor has worn off for a lot of users who found that ChatGPT is great for some things and mediocre at others.

Maybe this is OpenAI's way of saying: we know it's not perfect, here's how to get the most out of what it actually is. That would be refreshingly honest for a company that's spent two years promising artificial general intelligence is right around the corner.

Or maybe it's just customer education, nothing more. But what do I know. If you want to argue about it, my email's on the about page.

Sarah Williams · 57 mins ago · 4 min