Habr author warns: thoughtless AI use leads to specialist degradation

A Habr author raised the topic of what thoughtless use of AI could lead to: in his opinion, in the pursuit of profit, holders of large language models risk destroying not only the quality of texts, but also human qualifications.

Why Generated Text Does Not Replace Expertise

Generative transformers, according to the author, do not invent anything — they merely use large language models for generation, that is, combining plausible texts. Those who apply generated texts without proper critical analysis inadvertently lose qualification, if they had any at all. And those who were never qualified specialists fall into euphoria, imagining that now they can solve problems without possessing the necessary knowledge.

"This will work as long as truly knowledgeable people are steering the development.

But when they are replaced by those who have mastered interaction with AI, but do not possess the necessary qualifications to soberly evaluate results obtained with the help of AI — that's when disaster will strike," warns the author.

What Happens to Models from Version to Version

The author notes that from version to version, models add hidden deterministic mechanisms — what is often described by saying "the model thinks" — which should improve answer adequacy. But along with this, more and more human factor is embedded in model responses, and human bias can leave its mark on the final result. For commercial promotion of the product, this, as the author notes ironically, is certainly good.

A separate problem is the data sources on which models are trained: their accuracy is guaranteed by nothing. With code generation, according to the author's observation, things are somewhat better, but here too AI fundamentally does not invent anything new. At most, if models are trained on the results of generations recognized as most successful, some improvement or, conversely, degradation of final results can occur.

Who Is at Risk

The author's key idea is that the danger comes not from large language models themselves, but from how they are used by a person without proper qualifications. As long as decisions in companies are made by people with real expertise, capable of critically evaluating AI results, this problem remains manageable. The risk grows when such people in leadership positions begin to be replaced by specialists who can only interact with AI but are unable to soberly evaluate the quality of the result obtained. In essence, this is about a generational change within the profession: one generation still learned to solve problems by hand and mind, while the next risks learning only how to formulate requests to the model.

What to Do About This Risk in Practice

From the author's reasoning, a practical conclusion follows: the result of AI work should be viewed as a draft, not a finished answer, and checked by a person who is capable of evaluating the content on its merits, not just in form. This applies to both texts and program code — in both cases, the model combines what it has already seen rather than creating fundamentally new knowledge, and responsibility for the quality of the final result remains with the person who uses it.

What This Means

The author urges not to confuse the ability to use AI with professional qualifications: the tool combines plausible texts but does not replace the ability to critically check the result. As long as knowledgeable people remain in control, the risk is manageable — but it grows as they are replaced by those who have only mastered working with AI.