Let's Quiet Down About AI Already

Chapter Zero: Defining a Couple of Terms

Stupidity is when you don't know what was supposed to be there in the first place. When the key hook has been removed and you don't know about it — that's unpleasant. But it becomes "a very expensive and offensive kind of stupidity" when, for example, you pour diesel into a gasoline engine.

Something can only be called "working" if it works on its own, without supervision. Most machines require maintenance, but if before printing you need to "open a new application, press exit three times, then launch the old version" — that's not a working application. It's broken.

The author is a software engineer by degree, which implies the ability to "invent, create, maintain, and manage various programs and computer hardware."

The key point: A computer collects, stores, processes, and transmits information by executing a precise sequence of commands.

Chapter One: What Is Programming

Programming is an exact science. A specialist can "precisely tell you what will happen when a certain command is used" and evaluate the reliability of systems with a certain probability.

There are fault-tolerant systems, real-time systems (used in rockets and aircraft), banking systems with precise calculations. Computers repeat actions without changes indefinitely — that's their fundamental value.

Chapter Two: And Now, Ladies and Gentlemen — AI!

AI is "a big matrix multiplier that adds random numbers to its answers." The problem: this violates the very definition of a computer, which should "always produce an exact answer."

Investments in AI are enormous: an NVIDIA H200 costs $35,000 per card, and they're purchased 150,000 at a time. Electricity is a critical resource. In the US, investors are ready to pour 250 billion dollars into power plants by the end of the year.

The question: What's the point of AI? What's the return?

AI helps "process large amounts of information fairly quickly." But no model can guarantee a 100% correct result. "It needs babysitting" — this is not a working system.

A balance is needed: if you don't use AI, you'll be slow; if you offload too much to it, you'll "spend hours trying to fix what AI broke."

The danger: not knowing the difference between rm ./* -rf and rm . /* -rf can be fatal. "Not understanding and not reviewing commands produced by AI is the plague of modern development."

In banking development, the introduction of AI meets skepticism — and that makes sense.

Chapter Three: Who Actually Needs AI?

An accounting system runs on algorithms from the 1980s. "It won't be changed. It doesn't need to be changed." What matters is whether the system in 2025 produces the same answers as it did in 1990.

AI systems handle a limited range of tasks well: prediction and big data processing.

LLMs and text: "Rewrite this text for me, and replace all informal 'you' with formal 'you'" — works great. The result can be 99.95% correct. Then you just need to "sit down and proofread the text."

LLMs handle things well where a precise result isn't needed. Translations work if you don't need an exact translation — there will be "wrong shade words and idioms," but it's better than old automated translations.

Chapter Four: The Human Factor

Problems arise when hiring through AI. For every "no" from AI, "the person themselves has a perfectly reasonable justification or reason." AI might let through "the biggest jerk" or filter out a wonderful person who doesn't know by heart "the square root of an integral of the forty-second degree."

Conclusion

What they say: AI is a gold mine where you can't possibly lose.

Reality: AI is just another technology that needs to be learned, understood, and applied properly.

"Any idea, however wonderful it may be, will be ruined if used too little or too much." AI everywhere — stupidity. AI only in ChatGPT once a month — falling behind.

Don't fall for the marketing pitch. Better to ask for sales charts and revenue figures from companies that claim AI should be everywhere.