There is a new software technology. It has a bunch of boosters. There are some companies formed around the idea that they will make a lot of money by providing the technology. Then they go out and try to convince other people that they should use the technology – and by extension, pay them for it. If not them, then one of their competitors, but preferably them.

That’s marketing…

Marketing and advertising have a range of options available. The ethically best methods are effectively teasers for education: micro-lessons in what the product or service can do for you. Further lessons in why this is good. And then, product differentiation: why this one is better, faster or cheaper than the others. Technical products often get published reviews and benchmarks, which are a mixture of subjective and objective analysis. Sometimes they even measure things relevant to what you want to know!

The least ethical methods are lies. Things which aren’t true, or are deliberately deceptive, or deceptively irrelevant. Did you know caffeine is an aphrodisiac? It’s true: people are much more likely to want amorous activity when they are awake.

Somewhere in between are the social engineering methods. The big ones are FOMO - fear of missing out, that everyone else is getting benefits that you won’t - and FUD - fear/uncertainty/doubt instilled about the competitive products.

Testimonials can serve as social proof (that’s not the same as rational proof) that a given technology provided benefits to a particular person or organization. Or they can serve as social proof that other people are getting things that you are not.

If you see a new technology marketed mostly via FOMO, the odds are very good that it is fraudulent.

Blockchain (the technology) turned out to be entirely FOMO: nobody has found a legitimate use case for which blockchain is the best answer. Cryptocurrencies turn out to be good for gambling, fraud and similar borderline or outright illegal activities. All other proposed uses suffer from lack of an oracle relating reality to representation or by being more efficiently implemented by a traditional authenticated database.

Object-oriented programming turns out to have significant benefits in coordinating software built by large teams of programmers who don’t or can’t communicate efficiently. Since that describes a lot of large corporate software efforts, OOP has succeeded in the marketplace.

As of late 2024, LLMs and similar machine learning techniques – which have been successfully renamed by marketing as “AI”, despite not being anywhere near our cultural expectations of AI – have found niche success in translation and pattern continuance. Everything else that they are proposed for – and they are proposed for everything – is unsuccessful. The reasons are various: expense, speed, (in)accuracy, and lack of capability all show up regularly. In the normal course of events, we would expect LLMs to remain a niche technique, settling into a place in the programming toolkit not unlike regular expression pattern matching or declarative programming languages.

Unfortunately we are not in normal times, and LLMs are being marketed primarily by FOMO tactics: if you don’t adopt “AI” tech, you cannot be competitive with those who do. Nothing backs this argument up.

It may not be criminal, but it’s certainly not a good move in the majority of circumstances.