At least for businesses, public institutions, and government organizations.
The fact is that universal large language models (LLMs), as practice shows, are not the most suitable solution for performing specific specialized tasks. They do an excellent job as personal assistants, but as business tools they leave much to be desired.
To begin with, they are overly complex. Their use often resembles Goldberg machines — cartoonishly elaborate devices designed to perform simple actions like dropping a sugar cube into a cup or flipping a light switch.
During “marketing showcases,” audiences are dazzled by stories about massive databases covering every field of human knowledge used to train yet another artificial intelligence. And the magic of big numbers quite literally pushes company leaders and top executives to reach for their wallets.
But if we set emotions aside and assess the situation rationally, an inevitable question arises: what tasks truly require AI tools trained on billions and trillions of parameters? Calculating a flight trajectory to Jupiter’s moons? Designing a thermonuclear reactor?
When comparing the complexity of LLMs with their practical application, one can draw a disappointing conclusion: in most cases, they are used to “add two and two.” In other words, for the vast majority of tasks, their power is excessive.
In other words, by purchasing an LLM subscription, entrepreneurs and managers are often paying for capabilities they largely never use.
The second problem stems from the first. Complexity and universality give rise to the well-known phenomenon of machine hallucinations and reduced accuracy.
LLMs frequently make mistakes, “invent” facts, and cite nonexistent people and sources. In addition, LLMs tend to average and standardize solutions, which creates challenges in situations where creativity and personalization are required.
At the same time, it is worth noting that neither commercial companies, nor government organizations, nor public associations need to solve “all the problems of the Universe.” Each structure has its own specialization and, consequently, a specific set of tasks that must be handled with maximum quality. Therefore, the optimal solution for them is specialized AI tools, created and configured to work quickly and accurately within their specific domain.
Why should an AI accountant be able to write haiku or Shakespearean sonnets, and why should an AI legal assistant know the biographies of silent film actors? Any excess information beyond the purpose of a specialized AI tool increases the risk of errors and hallucinations. Conversely, narrowing an AI’s knowledge and skills to a specific specialization dramatically improves the quality and accuracy of its work.
This, of course, does not mean that LLMs are useless or should not be used. As universal personal assistants, they are arguably unmatched today. But specific tasks are better entrusted to “AI professionals.” Especially since more and more specialized AI tools are emerging every day — there are options in virtually every field of human activity.
By the way, if you need high-quality professional AI tools to create “live” avatars, “smart” chatbots, and video presentations with digital speakers — you know where to turn.
