AI Fails Basic Tasks, Sparking Debate on Utility

Debate over AI failures in basic tasks

The artificial intelligence promising to revolutionize employment cannot even count the letters in a word. In an extensive exchange of views, several participants have documented systematic failures in models like ChatGPT, Gemini, or HuggingChat: from inventing noun genders in Spanish to making elementary calculation errors. One participant summarizes it with irony: "It's ridiculous. You have to ask three times, explain where it goes wrong, and there's no way around it, even though algorithmically it is quite simple to program."

The initial discomfort is not an isolated case. Another contributor states that the tool has done "similar things" to them, although they admit that "it still surprises me how much it understands, even if it sometimes fails." The complaint extends to the lack of reasoning: "AI does not reason, it only repeats what it has been taught," says a third person, comparing it to a rigid civil servant. The metaphor of the chattering parrot runs through the entire discussion: a system that spits out probable answers without understanding what it says.

Why does AI fail at seemingly simple tasks?

The underlying problem is technical and acknowledged by developers themselves. A participant cites an explanation from ChatGPT itself: "LLMs generate predictable text based on statistical probability and linguistic patterns, they do not perform precise calculations or deterministic processes." That is, there is no algorithm that counts letters or assigns grammatical genders; there is a model that bets on the most probable word according to the context. Therefore, when asked for mathematical precision, it "improvises" instead of solving.

That probabilistic nature explains why a model can answer "two plus two" correctly ninety-nine times and fail the hundredth time with a "cherry pie." It is not a programming error, but the consequence of a system that does not calculate, but predicts. The paradox is that this same imprecision coexists with capabilities that astonish even skeptics. A user recounts how a veteran pediatrician was left "speechless" after verifying that the AI correctly calculated the dosage of a rare medication for a 15-kilo child, including the conversion from milligrams to milliliters.

Real impact on employment: between dismissal and productivity

While some debate technical limitations, others have already made moves in the labor market. A participant confesses frankly: "You can go tell several former employees I fired because a custom GPT chat performs the same repetitive tasks for €20 as opposed to €2,500/month, such as creating reports, analyzing data, or writing content." The figure is striking and summarizes the economic equation many companies are evaluating.

The replacement is neither total nor homogeneous. The same participant clarifies that AI "is not going to take everyone's job according to experts. It only takes jobs - for a mysterious reason - from code line programmers." Others point out that the tool saves time on specific tasks: preparing presentations that "look written by a City consultant," generating text drafts, or analyzing data. A contributor celebrates this: "For now, it has already saved me consulting fees, lawyers, community managers, and graphic designers on several occasions."

The gap between hype and everyday reality

Skepticism runs through the exchange like an underground river. "It is a system to appear to have a correct conversation, nothing more. There is no substance behind it," says a participant. Another compares it to automobiles: "Automobiles were not better than horses in their beginnings either." The discussion drifts toward the phenomenon of tech bubbles: "All tech trends are the same; after a few years online, you detect them without scratching deeper," notes a veteran who mentions graphene and drones as precedents.

The accusation of being a passing fad clashes with those who already use it daily. "We are actually learning new ways to interact with it. A year ago, I wouldn't have thought of what I do today with ChatGPT," replies a participant. The discussion gets heated when accusations of ignorance are exchanged and users block each other. Amid the noise, one reflection stands out: "The key is that it allows optimizing tasks and the time investment they require, helping to be more productive."

The problem lies with the asker, not the responder

One school of thought holds that most failures are not the machine's fault, but the user's. "Prompts must be very specific and logical, otherwise you are messing it up yourself," warns a participant. Another elevates this to a category: "There you have found the main limitation of AI: the idiot who doesn't know how to write orders for it." The idea that the tool is only as good as its operator gains trinc among those who defend it.

However, critics counterattack with concrete examples of failures that do not depend on the prompt. "Just ask it how many 'r's are in the word 'arrancar' [to start], and all models fail except OpenAI's 1 preview and another Chinese one," points out a participant. The inability to count letters or predict how many words its own response will have betrays the absence of a symbolic reasoning mechanism. For defenders, it is a minor detail; for detractors, proof that the emperor has no clothes.

The uncertain future of a tool that is already here

The discussion reaches no verdict. Some see a productive revolution underway; others, just another bubble. "AI is the future, and always will be," ironizes a participant, summarizing the skepticism of those who have heard unfulfilled technological promises for years. On the opposite side, a defender responds with a market argument: "Sure, dude, because it's going to stay like this forever and won't advance at all."

What emerges from the exchange is a powerful tool for certain tasks and clumsy for others, whose impact on employment is already materializing in some sectors while being exaggerated in others. Human supervision remains mandatory under EU regulations, an implicit recognition that reliability is not guaranteed. With these elements, the most honest prediction is that AI will continue replacing specific tasks—and some workers—without yet fulfilling the promise of a total revolution. The rest is, for now, smoke.