Even Gartner Got It Wrong. So No One Really Knows the Future of AI.

The other day, I was looking at Gartner’s 2023 Hype Cycle.

In that chart, large language models were already positioned as if they were starting to enter the Trough of Disillusionment.

In other words, the view seemed to be:

“Expectations were high, but now it is time to face reality.”

As we all know, that is not what happened.

Far from entering disillusionment, the AI industry accelerated even more over the next two to three years.

GPUs became scarce around the world, and major companies began lining them up by the hundreds of thousands.

Companies started competing over AI investment, and even governments began talking seriously about AI infrastructure.

According to the predictions back in 2023, prompt engineering was supposed to be one of the main themes around now.

It certainly had a huge moment.

But today, the conversation is much more about making AI think for itself, building agents, and having models call tools, rather than simply polishing prompts.

The speed of the trend cycle is almost absurd.

Of course, the Hype Cycle is a forecast.

There is nothing strange about forecasts being wrong.

Still, when it comes to LLMs, I think this was a fairly bold miss.

Even back then, many people understood that AI would become capable of writing reasonably natural text.

But few people expected it to become a general-purpose tool that could:

• Write programs
• Help design systems
• Teach English
• Offer life advice
• Summarize long documents
• Write research reports

Some researchers and developers were already talking about that potential, but I think many people still saw it as “a slightly smarter chatbot.”

So what happens next?

Honestly, I do not know.

Looking at the current trajectory, performance improvement has not stopped yet.

For the time being, we may keep adding GPUs, consuming more electricity, and moving toward a world where humans and AI compete for power.

At some point, we will probably hit physical limits.

After that, the direction may shift toward more energy-efficient GPUs, better algorithms, or achieving the same performance with far fewer computing resources.

…But after writing this much, I had another thought.

Even Gartner, one of the world’s leading research firms, could not fully read this wave.

So there is no reason to believe that my own predictions about the future will be right.

That is why, these days, whenever I see someone speak with certainty about the future of AI, I find myself thinking, “Is that really so?”

If next year the news says, “AI with a personality has now become practical,” I do not think I would be surprised anymore.

On the other hand, if someone says, “AI performance improvements have hit a ceiling,” I would probably accept that too.

When it comes to the future of AI, the most accurate answer right now is probably this:

“No one knows.”