AI won't steal your job: it still fails at multiplication

AI promises a PhD but still fails at multiplication

The conversation about artificial intelligence has moved beyond visionaries and tech profiles. The starting point is a concrete thesis: model pogre is exponential, and with upcoming generations — GPT-5 and GPT-6 are cited — machines would reach doctoral-level potential and surpass any human. From there, the question isn't if lives will change, but when and how fast. The most aggressive forecast speaks of months. The cautious one predicts a full industrial revolution with economic, political, and social effects, but also hype, snake oil salesmen, and competition.

What a language model really is

The first clash comes with the definition. Against the narrative of emergent intelligence, some analysis argues this isn't intelligence at all: they are predictive models responding based on probabilities and dependent on training data. The result can be spectacular, but the gap with real intelligence remains huge.

The test proposed is simple and devastating: ask for a multiplication of three digits or more. Or ask what the fifth word of the first paragraph of the second chapter of Don Quixote is. Repeat the operation several times and compare. That's where the myth falls, they say, because the invention doesn't calculate: it approximates.

The oracle that doesn't think and the voter who applauds

The second focus is social, and the most uncomfortable. If AI becomes the default source of answers, the effect on a population that no longer seeks links but accepts direct responses is evident. It is described as an oracle: whatever AI says is truth, period. And thus, whoever controls the oracle controls the narrative.

Some take it directly to politics: given the average voter's level, AI can be whatever its handler wants it to be. Things are added or removed as convenient, and citizens applaud without thinking. The repeated conclusion is that the key is for citizens not to think.

Economic impact: productivity and replacement

In strictly economic terms, the diagnosis is more nuanced. The immediate effect isn't massive replacement, but the gap between companies that prematurely adapt technology to their workers — with brutal productivity gains — and those that don't, which will be devoured by the former. Worker replacement, however, would be affected sooner than expected.

The comparison used is that of a new steam locomotive, not a new video player. An industrial revolution with all its changes. And with it, the healthy effect of competition: only the most productive companies would succeed, at least in theory.

The practical limit: it works if you know how to use it

Real-world usage experience significantly lowers the hype. The tool is excellent as a quick and reliable assistant around 80% of the time, but the other 20% requires knowing how to prompt, detecting inconsistencies, and insisting when the answer is vague or superficial. Without solid knowledge of the subject, it's worthless.

The nuance matters: it doesn't replace the expert; it amplifies the expert. And here appears the usual gap, separating those who use the tool from those used by it.

The bubble, the gold rush, and the underlying collapse

The most cynical reading is financial. Behind the magnification lies something more prosaic: sustaining the collapse of an economy needing a new golden age to maintain faith, a narrative allowing further expansion to cover debt fraud. When the bubble bursts, this school says, people will rush to talk about another trend.

This isn't unanimous. Opposing it is the evidence that the internet also had its bubble and is now everywhere. Denying AI's utility would be short-sighted. There will be hype and pruning, but the useful parts will remain.

Usage already changing things

While the future is debated, some are already using it. Image creation, video from images, voices, avatars, comics, personal stories with quality unthinkable recently. Pseudo-humans at will, with voice and life, and social media interaction. AI as a creative tool for those without teams or budgets.

And on the dark side, cyberfraud. The ability to copy image and voice — even speaking other languages — to the point where a mother wouldn't recognize the fraud. This is the usage that is already here and doesn't need to wait for superintelligence.

What remains unresolved

The question of whether these systems learn on their own or just adapt to the user remains open. It is pointed out that many models are already trained on AI-to-AI conversations and it's only a matter of time before they learn from themselves. That would be the step making evolution exponential. Not tomorrow or in a year: it's a matter of time.

And in parallel, the fundamental doubt: if knowledge is overwhelming but inventiveness is zero, if it's brute force and not creation, where exactly is the leap? The debate doesn't close. It stays right there, at the point where data and promise don't quite fit.