Hype is a Hell of a Drug

Confused? Come on in! Join my self-help group for recovering hype junkies as I explore the Dangerous Chemistry of Storytelling and share my personal skepticism on AI's wild ride right now.

My friendly Roman Francesco edited this edition and helped turn a rant into, well… you decide!

---

---

With every fresh declaration, I couldn’t help thinking that it sounded like complete and total bullshit. Every update was met as if it was the second coming of Christ. It’s been going on for just over 2.5 years and while the technology has come on leaps and bounds, it is still limited and the improvements, to me at least, were becoming less profound.

I’m of course talking about AI. For the past six months or so, OpenAI’s executives have been talking about artificial general intelligence as if it were literally around the corner. An extremely near-term inevitability.

Sam Altman himself described GPT-5 as “a significant step along the path to AGI”. As an ex-tech journalist and PR pro, OpenAI’s PR team executed the game plan almost to perfection. Hints, leaks, and carefully stoked expectations painted a picture of GPT-5 as the moment the company would cross some invisible line between powerful tool and proto-AGI.

The campaign only fell down in one place. Unfortunately, it tripped up at the most important part. The actual product itself. And the problem with the kind of hype narrative stirred up by OpenAI is very simple: if you don’t deliver, you disappoint.

Benchmarking Reality Check

For balance, it is better. It’s faster and smarter. For example, it scored 74.9% on SWE-bench Verified compared with GPT-4.1’s 54.6%, according to data from Leucopsis. Cornell University researchers found that in medical reasoning, it improved reasoning and understanding by 29.6% and 36.2% respectively over GPT-4o.

GPT-5 even shows promise in mammography image interpretation, however it still lags trained clinicians in some areas of that field.

Don’t get me wrong, this is all great stuff but for all of GPT-5’s polish (and there’s plenty), it failed to meet expectations. It’s not a leap towards AGI. It’s barely more than a pigeon step towards AGI. Don’t take my word for it.

“GPT‑5 failed the hype test,” writes The Verge, noting that despite grand promises, users reported only incremental improvements and found the model “underwhelming” and “not a massive step forward.”

It’s incremental, not transformational. The narrative was akin to landing humans on Mars, but instead we’ve got a software patch. A capable upgrade and nothing more.

The Inverse Exponential: Fireworks First, Sparks Later

Privately, I’ve long argued that AI’s trajectory is wrong. The narrative has suggested endless exponential gains, yet reality has been the exact inverse. The big leaps were early and now are being followed by smaller hops or even nudges.

Like fireworks, the biggest, brightest bursts come first, followed by smaller pops that still light the sky but no longer dazzle in the same way.

GPT-5 is another spark, not the grand finale. Mobile phones weren’t indispensable overnight. They became mainstream through hundreds of tiny and boring iterations by Swedish, Finnish and Canadian companies. Even when smartphones arrived, they exploded on the scene and the improvements like better batteries, sharper cameras and faster chips aren’t lifechanging.

AI may now be entering that same phase.

The Cost of Over-Storytelling

The danger of overselling is that it erodes credibility. If every new release is billed as “a step toward AGI” and you don’t deliver, the credibility gap widens. Gartner calls this the “cycle of cynicism.” Once customers, regulators, and investors feel burned, they stop evaluating each claim on its merits and start assuming all claims are inflated.

The problem with the credibility gap is that it has a long tail. In consumer markets, customers churn by declining upgrades or downgrading to free tiers. In B2B, execs set tougher benchmarks or ask for longer pilot projects. In capital markets, it kills valuations or, worse, leads to ferocious corrections when everyone agrees that it’s overhyped.

For OpenAI and its peers, that’s not just a PR stumble. It’s a strategic risk. Trust is far harder to rebuild than it is to maintain. Microsoft learned this in the 1990s after over-promising on early Windows launches. Tesla has learned it repeatedly in autonomous driving claims. It’s even worse for the leading LLMs. Realistically, 99% of use cases are exactly the same, so OpenAi, Anthropic et al won’t generate blind brand loyalty (like Apple), so they don’t have a die-hard fanbase (unlike Tesla).

Ironically enough, the quantum industry has long been accused of being overhyped and every quantum company I’ve ever met is at pains to be as factual and sober as possible. I spend a lot of time advising quantum startups to show off a little ankle and flirt more.

The Investment Paradox

Yet in AI, the money keeps flowing. In the first half of 2025, AI startups pulled in 53% of global venture capital funding, and 64% in the U.S., according to Axios Pro Rata. More than half of every VC dollar on earth is now chasing AI. It’s distortion.

OpenAI itself raised $40 billion at a $300 billion valuation this spring and yet, Sam Altman, architect of the hype cycle, has been unusually candid: “AI is in a bubble… someone is going to lose a phenomenal amount of money”.

This is the paradox: investment is peaking just as narrative credibility is slipping. The checks are still being written, but the stories behind them feel shakier. You don’t often see this much capital sprinting into a sector just as the sector’s leading figures admit the hype might be outpacing the fundamentals.

It feels eerily like the dot-com bubble: money flowing not despite skepticism, but because investors fear missing the next Amazon. However, back then the storytelling hype was far less enthusiastic.

The Real Lesson: Storytelling Discipline

So what’s the takeaway? AI isn’t entering the trough because of its actual capabilities. They are real, and improving. It’s entering the trough because of the way those capabilities were framed. GPT-5 wasn’t bad. Far from it. It was incremental, dependable progress — arguably the kind of progress enterprises need most. But the story told was bigger than the product could carry, and the fallout is that trust takes the hit.

This is the lesson for founders, investors, and technologists: storytelling is leverage, but it’s also liability. Hype can bend markets and pull forward adoption, but if it overshoots, it plants the seeds of disillusionment. And when storytelling outruns reality, reality always wins.

Better to promise steady progress and surprise people than to promise AGI and deliver a service pack.

Gartner’s Hype Cycle: Sliding Into the Trough

This is why Gartner’s framework resonates so strongly right now. The 2025 Hype Cycle places Generative AI squarely in the ‘trough of disillusionment’, according to Gartner. Even The Economist put it bluntly: “Many bosses are sliding into the ‘trough of disillusionment’”.

The irony is that GPT-5’s incrementalism is exactly what we should have expected, which is a tool getting steadily better, a platform refining itself. A trajectory toward usefulness rather than transcendence. But because the story was overplayed, the trough feels deeper.

When Hype Meets History

And this isn’t the first time we’ve seen the script. Virtual reality was going to replace the office. Blockchain was going to replace the bank. Autonomous cars were going to replace the driver. Each promised a revolution; each hit the wall of complexity.

The pattern is eerily consistent: the initial leap dazzles, early adopters rave, capital floods in — and then the technology stalls on the hard edges of physics, cost, or usability. AI isn’t immune. GPT-5 doesn’t mark failure, but it does mark normalization. The shine wears off, the story gets less sexy, and the real work begins.

The Plateau of Productivity Is the Prize

The hype cycle doesn’t end in despair. After the trough comes the “plateau of productivity”, which is the point where technology matures, becomes boring, and quietly powers the world. Cloud computing is the obvious example: no longer hyped, but utterly indispensable.

GPT-5 may be less a letdown than the beginning of that plateau. A shift from dazzling demos to dependable tools. The irony is that the plateau is often where the real money is made. AWS wasn’t built at the hype peak of the cloud; it thrived when the hype was gone and the boring work began. GPT-5 might represent that same inflection point for AI.

Where Next?

The good news is that this doesn’t spell collapse. It spells maturity. If the exponential dream was always a myth, then what’s left is the durable, steady growth of a technology finding its real place.

GPT-5 isn’t AGI. It’s something more pragmatic: a plateau of productivity, where AI is less about headlines and more about infrastructure.

And maybe that’s the real story worth telling.