Hype Is a Business Tool

Remember WAAAY back in late 2022 (what feels like ancient history now) when you first started playing with ChatGPT? Like everyone else, you probably created a poem in a pirate’s voice. “Pirate Poetry” was fun, exciting, and even playful.

Today, EVERYONE is talking about AI, and the conversation is all over the map. You have loud boosters claiming their lives have been completely transformed (just buy my book!). You also have haters that despise everything about it. When I mentor UX designers, many tell me they’re worried and feel like their careers are about to be stolen from them.

First, let’s pause just a bit. Can we please stop calling it “AI”? It’s much more accurate to say LLMs (Large Language Models) but so many have just settled on AI. I’ll actually be “that guy” and stubbornly use LLM instead of AI throughout this post just because I can’t be party to this oversimplification. Don’t worry, when I want to be snarky, I’ll use AI.

I’m constantly trying out various LLM tools and find they often perform far poorer than promised. When I try to discuss this online the boosters come out in great force, often with a dismissive response about “you’re prompting it wrong”. For a UX designer, blaming the user is as close as you can get to a cardinal sin. But this boosterly vibe seems to have escaped social media and is infecting the C-Suite of many companies. Cory Doctorow said it well: “we’re nowhere near the point that AI can do your job, but we’re past the point where an AI salesman can convince your boss it can.” For such a young and still unproven technology, it’s shocking how quickly it’s being forced upon teams.

I want to be clear, I’m not saying this tech is doomed. It clearly can do some impressive things. My point is that the hype is running too far ahead. Instead of a more nuanced discussion about how it can be used, polarized camps have formed yelling at each other. I’d like to cut through some of this noise, to offer a more grounded perspective on LLMs, and to explore how UX principles can serve as our anchors in this hype-infested online discussion. This post is part 1 in a series about LLMs. I’m starting with helping us talk about LLMs. How we talk about this, both in person and online, is so infested with hype that it’s hard to have a genuine conversation. It’s so easy to be confused and dejected.

The Hype Machine and Our Collective Amnesia

I know this is going to make me sound old, but to understand where we are with LLMs, it helps to understand a bit of history. We’ve seen this same pattern many times, where a new technology comes along and everyone thinks it’s going to change EVERYTHING: quadraphonic stereo, 3D printing (someone once told me that in only a few decades, we’d be 3D printing iPhones), VR/AR, 3D TVs, and, more recently, the speculative bubbles of crypto and NFTs. This “love it!/where is it?” cycle isn’t new. As Mark Twain said, “History doesn’t repeat itself, but it often rhymes.”

This cycle has a name: the Gartner Hype Cycle. It’s well known in the tech industry but I’m surprised by how many UX people I talk that to have no idea what it is. It has five distinct phases:

Innovation Trigger: A breakthrough technology emerges.
Peak of Inflated Expectations: Early publicity, success stories, but also numerous failures.
Trough of Disillusionment: Interest wanes as experiments and implementations fail to deliver.
Slope of Enlightenment: Understanding grows, and realistic applications emerge.
Plateau of Productivity: Mainstream adoption and real-world relevance.

What drives the early part of the hype cycle is usually venture capital. They don’t want to be left behind when some new hyper growth technology appears so the money flows rapidly. This funding generates lots of startups, online speculation, articles and blog posts. Everyone begins talking about this new thing and it starts to feel inevitable, until it doesn’t. It’s a story as old as AOL.

MOOCs: A Perfect Example

MOOCs (Massive Open Online Courses) are a perfect example of a hype cycle. They started gaining popularity around 2011, quickly leading to the founding of companies like Udacity and Coursera.Venture Capital poured in, with promises that MOOCs would completely revolutionize online learning to the point where it was predicted “In 50 years, only 10 U.S. colleges would remain”. This type of hyperbole sounds ridiculous today. However, a few years later, the bubble popped as it became clear that completion rates were very low. The initial vision faded quickly, leading to a pivot towards something much more modest: training certificates.

Why does this happen? Most new tech starts off as naive. John Seely Brown discussed this in The Social Life of Information, where he documented how companies in the 1980s, when moving from paper to electronic databases, naively replicated only the text fields. They failed to appreciate the nuances of how people actually used the physical paper: notes in the margins, bent corners to mark a page, and which document was on top of the pile to indicate importance.

The technology wasn’t wrong, but the initial approach was simplistic, failing to grasp the deeper human behaviors that surrounded its use.

Mobile Went Through It Too

But I can hear you muttering mobile didn’t follow this cycle as it far exceeded expectations. The truth is, even mobile phones went through the same hype cycle. I should know; in 1998, I was working at Symbian in London when Europe was the epicenter of the mobile revolution. Nokia was churning out amazing handsets, SMS was freaking out parents all over the world, and monthly bills regularly tipped over £200.

Siemens first WAP phone

When WAP was first announced, people “just knew” a digital revolution was coming. The first Siemens smartphone, with its two-line display (😱!!), was supposedly going to let you surf the web on your phone. Just like MOOCs, it was a reasonable idea held back by ridiculously naive assumptions. Those early phones failed massively, and consumers backed away.

But notice that the mobile revolution didn’t die. It had its “trough of disillusionment” but kept slowly improving. It took a few years, but in 2007, the Nokia N95 emerged as the first app-based phone with a reasonable screen, which was quickly followed by the initial iPhone.

However, people forget the iPhone did NOT do very well when it first launched. It was laggy, it lacked apps, and it had painfully slow data. I was managing the mobile UX group at Google at the time, and we were desperately trying to make the original iPhone do something interesting. It wasn’t until the iPhone 3GS arrived two years later, with its massively faster processor and 3G data, that the mobile market finally exploded.

Mobile followed the hype cycle, peaking in the early 2000s, retrenching for a few years and slowly climbing out of the trough of disillusionment until it finally took off around 2009. The fact that it eventually exceeded the original peak of inflated expectations is what people remember the most. They forget it too had a crash of expectations.

Hype is a Business Tool

The key point here is that nearly every tech innovation has gone through this cycle. Note that it doesn’t mean that everything fails, it just means that everything has a fall after an initial naive peak. Occasionally, technologies like Mobile far exceed the original peak but that is the exception to the rule.

The reason this peak consistently happens is simple: hype is a business tool. Companies like Theranos, Udacity, Tesla, and now OpenAI understand that the money will eventually run out. They know they’re running on borrowed time.

They pump things up, pushing and promising, to secure as much funding as possible before the inevitable bubble bursts. This is why they make outlandish claims like “we are afraid of GPT-5” or “most jobs will disappear.” These are manipulative comments intended to freak you out, and they exist only to keep the money flowing for as long as possible.

But this massive inflow of money just doesn’t add up. The costs of running LLMs far far outrun what they are currently charging for them. At some point, people will realize the pitiful inflow of cash just doesn’t cover the costs. This is why the hype is so helpful. OpenAI is desperately trying to run negative balance sheets long enough to find something huge to pay the bills.

I’m not saying LLMs are doomed, I’m saying don’t freak out. It is VERY likely there is going to be a trough of disillusionment with LLMs. Will it be followed by an even bigger peak like mobile or crash like Crypto? That’s impossible for anyone to predict. But the technology is clearly being naively used and multiple studies have shown that many companies are having a hard time making their LLM projects actually work. This mirrors what happened with early mobile web pages and mobile apps. It takes a lot of mistakes to figure out what really works.

Clearly there is momentum here. Google, Facebook, Microsoft, Nvidea, OpenAI (et al) are all making a huge play here and it feels naive for lil’-o’-me to be saying they are doomed. But I’m not. I’m saying they are part of a classic hype cycle, one built on overly simplistic assumptions and things are very likely going to pop, at least in the short term.

People who understand this tech, its strengths and, more importantly, its weaknesses, are the ones likely to climb the “Slope of Enlightenment” and actually build something useful. This approach isn’t based on hype; it’s a pragmatic, bottom-up process of building and learning. Instead of swinging for the fences, you attack smaller, even boring, problems first. It may not be where you finally end up, but you’re building on a solid foundation of learning.

The path to genuine progress comes from building from the bottom up, not from hype down.

Next week I’ll post part 2 where I discuss how we anthropomorphize LLMs far too much and this blunts our ability to understand how to use them.