The theory of the flying rock

People love to give technology or technology solution names which makes it sound better than anything else. Do you remember the terms like Smart Cities or Smart Phones? Looking back, I don’t think any city since then became extra smart. And I don’t even know how to measure the smartness of my phone.

The same thing is now happening with AI — Artificial Intelligence. It became a synonym for Large Language Models. Ambiguous and ubiquitous at the same time. This nebulosity allows people to attach unreasonable expectations to technology which is not capable of fulfilling all these dreams.

It reminds me of the theory of the flying rock. It goes like this. Imagine a sunny day on the beach. Somebody finds a nice rock and throws it into the air. The rest of the party notices only the flying rock and are fascinated by that. Without asking a basic question — Can a rock fly? — they start discussing all the possibilities of what can be done with a flying rock. Some suggest that there have to be rules governing flying rocks, while others discuss the need for the responsible management of flying rocks.

Before you know there is a whole industry around flying rocks, new associations are formed and certifications established and suddenly the world is full of experts. Anyone questioning the rock’s ability to fly is relegated to the role of a technological Neanderthal.

It has to be noted that from time to time we have to help the rock to keep flying, but it is only temporarily before the necessary improvements are made.

We are nearing the moment when both Anthropic and OpenAI are going public. Both companies are aiming for a valuation of close to a trillion dollars. I am the last one to try to predict the success of this event. What I know is that neither company ever addressed the fundamental issues underlying both Claude and ChatGPT. Both systems are Large Language Models (LLMs) and any output is a statistical output — for the same question, you get a different answer. And the accuracy of these answers is not known since the accuracy of the training data is not known.

CEOs of both companies talk about the amazing future and what the technology will be able to do and the impact they will have on society. The interesting thing is that as they are getting closer to the IPO, they are reversing their own predictions — especially around the displacement or disappearance of jobs . In order to generate the income they need to justify the valuation, they need companies to use as many tokens (token is a unit of text which the LLM uses to parse an input or create an output) as possible. Naturally, there is a limit of how many tokens per day a person can use. Yes, there are examples where developers burn through millions of dollars worth of tokens. But that’s only one segment of the market.

Should your argument be that this segment can generate billions of dollars, I would like to remind you that these companies will need hundreds of billions of dollars in revenue to justify their respective valuation. Which means their technology has to be used by millions and millions of people paying for it. And paying a lot.

If that’s the case, either the technology has to be so advanced and reliable that it allows an existence of one person company (or does it really need a person?); or it will require as many people within companies to use it.

Supposed the technology is — or will become — so good that one person can run several businesses on their own. If that’s the case, why would OpenAI or Anthropic sell this to the public? These two companies could take over any market, any industry with their own technology. And since they don’t, it suggests that what they have is an impressive piece of technology which they will try to sell to companies in massive volumes.

Suddenly, the term AI becomes just a marketing term and companies will start evaluating the technology for what it really is. That will be the moment when companies start talking about measurable outcomes and being able to justify the spend.

This whole era reminds of the year 2000 when customers were giving me a hard time when they realized that the version of software they were running was a few weeks behind the latest release. Their objective was to run only the latest and greatest. You can imagine what it was doing to productivity and how expensive it was to keep upgrading all the servers and desktops. It was not sustainable. The dot com crash brought everyone to their senses and companies started focusing on what really mattered — the business.

Going public will force both companies to focus on the business side of things. Their customers will demand stability and predictability. Releasing a new version of their models every few months will not be sustainable for the vendors nor desired by their customers. And all that is assuming that the technology is not a flying rock.

Introduction of new technology follows the same pattern, this one is magnified by the amount of money invested in and the hype surrounding it. Other than that, it is the same. We are now waiting for the applications where this technology will be used and deliver benefits. You can find similarities with laser technology. What was meant to be used with sharks, ended up as a toy to amuse cats.

The recurrent pattern? Recognize when the technology is real and when it is a flying rock.