The Generative AI Hype Cycle: Why Most Projects Fail and What to Do Instead

It’s officially “that” year again—you know, the one everyone’s been buzzing about. The year is twenty-twenty-four, and it seems like everywhere you turn, someone’s got a hot take on generative AI. And why not? It’s pretty much taken over the world, with a whopping eighty-seven percent of companies jumping on the bandwagon to build their own applications. That’s right, eighty. Seven. Percent. It’s like everyone and their grandma is trying to get in on the action.

But here’s the catch: a whole bunch of experts are saying that most of these projects are, well, gonna bite the dust. It’s like trying to build a rocket ship out of toothpicks and glue—ambitious, sure, but also kind of destined for disaster. Why the gloomy outlook? Well, it seems like everyone’s gotten a little carried away with the whole “generative AI is magic” thing. The truth is, it’s not as simple as typing a few commands and watching the magic happen.

“Vibes” Don’t Make for Good Data Strategy

Here’s the thing about generative AI: it makes skipping the hard stuff way too easy. Data preparation? Model training? Who needs ’em, right? Just throw some prompts at the AI and boom, you’ve got instant results—or so it seems. This whole “prompt engineering” craze has everyone thinking they can hack the system without putting in the actual work.

But hold your horses, buckaroo. Without a solid data strategy—you know, the kind that involves actually understanding your data and making sure it’s squeaky clean—you’re basically flying blind. It’s like trying to bake a cake without a recipe and hoping for the best. Sure, you might get lucky, but chances are you’ll end up with a big, ol’ mess.

Shreya Shankar, a machine learning engineer, puts it perfectly: “Without proper data preparation and evaluation, expectations are set purely based on vibes.” And let’s be real, “vibes” don’t exactly scream “reliable data strategy.”

The Missing Ingredient: The Hard Work of Constant Tuning

Let’s talk about traditional machine learning for a sec. You know, the kind that’s been around for a while? Well, in that world, success doesn’t just happen overnight. It takes constant tweaking, like a mechanic fine-tuning a high-performance engine. You gotta keep an eye on your models, inspect your data, and make improvements along the way. It’s a whole process, people!

But with generative AI, it’s like everyone’s forgotten about the importance of hard work. The ease of use can be deceiving, leading folks to believe that they can just set it and forget it. But here’s the kicker: generative AI can be kinda, well, finicky. The relationship between the prompt and the response isn’t exactly a perfect science. You might give it the same prompt twice and get two completely different answers. It’s enough to make your head spin!

The Teenage Analogy: Managing Unpredictable AI

Amol Ajgaonkar, the CTO of product innovation at Insight, has a pretty hilarious way of describing how we interact with large language models (LLMs). He says it’s like trying to have a conversation with a teenager—you know, full of mood swings and unpredictable responses. One minute they’re giving you insightful wisdom, and the next they’re rolling their eyes and muttering, “Whatever.”

And he’s not wrong! Generative AI, for all its power, can be just as inconsistent and frustrating as your average teenager. You can feed it all the right information, phrase your prompt perfectly, and still get an answer that’s completely off the wall. It’s enough to make you want to pull your hair out!

The point is, we need to adjust our expectations. Generative AI can do amazing things, but it’s not a magic genie that can grant all our wishes. It’s more like a powerful tool that needs to be handled with care, patience, and a healthy dose of humor.

Starting Simple and Prioritizing Understanding

So, how do we avoid getting swept up in the generative AI hype and ending up with a bunch of failed projects? Well, for starters, we need to take a step back and ask ourselves a fundamental question: Do we really need generative AI for this?

You heard that right. Before you go diving headfirst into the world of LLMs and fancy algorithms, maybe—just maybe—consider the simpler options. Sometimes, a good old-fashioned heuristic or rules-based approach is all you need.

Santiago Valdarrama, the founder of Tidyly, is a big proponent of this approach. He stresses the importance of thoroughly understanding the problem you’re trying to solve before you even think about throwing AI at it. Often, he says, you’ll find that a simple set of rules can get you eighty percent of the way there. And the best part? It’s way less complicated and less likely to blow up in your face!

Think of it like this: If you’re trying to build a birdhouse, you don’t need a team of engineers and a supercomputer. A saw, some nails, and a basic understanding of carpentry will do just fine. Save the high-tech tools for when you’re building, you know, an actual spaceship.

When to Consider Generative AI and How to Approach It

Okay, but what if you’ve exhausted all the simpler options and you’re convinced that generative AI is the way to go? Well, even then, it’s crucial to approach it with a healthy dose of caution and a commitment to continuous learning.

Instead of going all-in on the most complex model right off the bat, start small and gradually increase the complexity as you gain more experience. Think of it like climbing a mountain: You wouldn’t start with the most treacherous route, would you? You’d start with the easier paths, learn the ropes, and gradually work your way up to the more challenging climbs.

And remember, even when you do decide to embrace the power of generative AI, it’s essential to manage your expectations. It’s not a magic bullet, and it’s going to take time, effort, and probably a few headaches along the way, to get it right. But by focusing on understanding, starting simple, and embracing the iterative nature of AI development, you’ll be well on your way to harnessing its true potential.