AI: GENIUS OR ARTIFICIAL IDIOCY?

Contrary to the promises that have been made, what is currently marketed to us under the name of AI (Artificial Intelligence) is, rather, closer to “Artificial Stupidity.” And the more clearly people come to understand this, the more effort the AI industry makes to render us dependent on it, so that it can soon demand an exorbitant price for its use. And it seeks to do so as quickly as possible, as bankruptcy is looming on the horizon. Let us be clear: we do need this technology, but not under the conditions presented by the AI industry and the so-called “smart money” surrounding it. In what follows, we will explain the games, the manipulations, and, more generally, what lies behind what you are being told about AI. We are Angelo and George, with 19 years lived under Ceaușescu, plus more than two decades of professional experience as a lawyer (Angelo) and, respectively, as a teacher (George).

2023: AI enters the game

AI made a striking public debut in 2023. Until then, known mainly to specialists and science-fiction enthusiasts, artificial intelligence suddenly became omnipresent in our lives, largely due to Sam Altman, CEO of OpenAI and creator of ChatGPT. In launching this program/application, Altman made bold statements suggesting that ChatGPT and AI in general would turn everything upside down and transform our lives forever. This was followed by what appeared to be a “coup d’état,” accompanied by the dismissal of Sam Altman from his positions. The company’s Board of Directors sent him home, with the remaining board members allegedly alarmed by the prospect of an artificial intelligence being created that could escape human control. A few days later, there was a dramatic reversal: the “people” — that is, the company’s employees — rebelled, removed the “junta” that had ousted Sam Altman, and reinstated him with even greater powers than before. A genuine Hollywood-style melodrama unfolded, widely covered in the media and on social networks. The spectacle was complete, and so were its consequences: ChatGPT and AI in general made a triumphant entry into our lives, provoking admiration and amazement. Almost everyone now sees AI as a miracle of the 21st century, a kind of 2.0 magic (the kind produced by the Internet and social media).

2025: ChatGPT versus Claude

For two years, until 2025, ChatGPT was the king of AI; in most contexts, “AI” essentially meant ChatGPT. But at the beginning of 2025, its position was challenged by Claude, developed by the company Anthropic under the leadership of Dario Amodei, which aimed to dethrone ChatGPT and take its place in users’ preferences. The offensive began with two advertising spots claiming that Claude was a better AI than ChatGPT. These ads were broadcast during the Super Bowl (the final of the American football championship), watched by hundreds of millions of viewers in the United States and around the world. A merciless war then began between the two AIs (more precisely, between the companies behind them), now competing for the same prize: investors’ and users’ money. Mutual attacks multiplied, endless debates followed, and successive releases of supposedly superior versions were presented as breakthroughs over the competitor. The goal was clear: larger funding rounds and a stronger market position than the main rival. Underneath it all was the attempt to convince users that only these systems existed on the market — that there were no other AIs beyond theirs. This, however, is far from the truth, as many alternatives exist in reality.

The discreet actors of the AI industry

Numerous companies produce artificial intelligence systems, but their products are far less visible in the public space. Some are smaller firms lacking the capacity for large-scale visibility — such as Perplexity in the United States or European companies like Mistral AI in France — while others struggle with promotion and user outreach, as is the case with Chinese AIs such as DeepSeek and others. There are also systems such as Grok by Musk, Gemini by Alphabet (Google’s parent company), Copilot by Microsoft, or Llama by Meta (Facebook’s parent company). Their creators have, in many respects, chosen a more discreet posture. They allow ChatGPT from OpenAI and Claude from Anthropic to occupy the spotlight, to compete publicly, and to present AI as something indispensable to modern life. In short, this is a kind of “free-rider” strategy in the broader media noise: benefiting from the enthusiasm generated by the two main actors without engaging heavily in marketing expenditures or assuming public controversy. It is, arguably, a rather “intelligent” strategy, given the current dynamics of the field and beyond.

Intellectual waste on an industrial scale

Today’s AI technology does not live up to the promises made about it. It is not possible to obtain high-quality work using AI without substantial input from the user. In fact, the AI systems available today — at least those accessible to the public — tend to present themselves as all-knowing and all-capable. In theory, they can do anything, at any time, and better than humans. In practice, however, when left on their own, they excel at producing what can only be described as large quantities of intellectual waste — especially whenever the user does not actively work alongside them. User disappointment is growing steadily, as AI companies and the “smart money” around them continuously present newer and supposedly more advanced, more capable, more “responsible” versions. Yet, when used in real-world conditions, users often find that each new program, application, or version is no better — and sometimes worse — than the previous one. It is also increasingly easier to “jailbreak” them (i.e., bypass usage restrictions intended to prevent illegal and/or unethical use of AI).

The “gibberish language” of AI

It is constantly present in AI-generated texts. Such outputs are easily recognisable due to their stereotypical phrasing, artificial and rigid language, lack of imagination, and repetitive reliance on cliché patterns that appear mechanically assembled — as if produced by a jaded engineer with the creativity of a rusty bucket. To further persuade users and reinforce the illusion of progress, AI companies have begun introducing so-called specialised units for specific tasks: programming, logistics, legal work, human resources, and so on. These are now called “AI agents” to distinguish them from “AI models,” which are supposedly the previous generation. Of course, all this is accompanied by definitions written in the same polished corporate jargon — the only domain in which AI consistently excels. For example, Gemini (Google’s AI) describes an “AI model” as a mathematical algorithm trained on vast amounts of data to perform a specific task such as price prediction, text generation, or image creation. An “AI agent,” meanwhile, is defined as an autonomous entity — a superior computational system that uses one or more AI models as a reasoning engine, supplemented by memory, tools, and the ability to interact with the external world to achieve a global objective. And thus, the “AI revolution” reveals itself, in reality, as a “marketing revolution for AI” — or, more precisely, a “sales revolution of AI.” Everything is done to convince us that an “AI agent” is something different, more advanced, and more “intelligent” than an “AI model,” so that we move on to it — and pay more for its use than before.

The Rabbit Out of the Same Hat

The so-called “AI agents” that began appearing toward the end of 2025 are nothing more than the “AI models” of 2023–2024, repackaged as something supposedly new and more advanced (nonsense, frankly). In essence, it is the same AI technology, the same “AI models,” merely “trained” and primarily fed with domain-specific data for their so-called “specialisation,” and then given a new name. A sleight of hand trick — and voilà! A new generation of AI technology is announced. It is therefore not surprising that the results delivered by these “AI agents” are often disappointing, far from the performance advertised by their creators. And it is very likely that next year we will see another round of triumphant announcements in the same vein. However, as this strategy begins to lose credibility, others are now attempting to develop genuinely new AI technologies.

Between Simulation and True Intelligence

One example of a possible new direction in AI is the “World Model” proposed by Yann LeCun (former head of AI at Meta): an AI system that contains within itself a simulated representation of the environment in which it operates, including its state and dynamics. Within this internal space, the AI runs simulations for the tasks it is given: it tries different approaches, tests outcomes, and evaluates their potential consequences. The final output is only delivered once internal simulations indicate that the expected level of quality has been reached. Whether this will be feasible, and whether it will lead to genuine “artificial intelligence,” only time will tell. For now, we must note that the king — or rather the queen (since artificial intelligence is grammatically feminine in several languages) — is naked, or more precisely, devoid of… intelligence.

The Limits of Algorithmic Creativity

At scale, it quickly becomes evident that AI is mainly useful for two things. The first is providing information and/or texts already contained in its dataset — in short, a more advanced version of Google Search. The second is “blending” existing texts, data, and information: when given a task, AI retrieves relevant material from its database, mixes it into a more or less sophisticated amalgam, and adapts it to the prompt. In essence, this is exactly what we could do ourselves — provided we have the time, the willingness, and the intellectual effort required. In general, the output produced by AI is of clearly inferior quality to what a human could achieve, for a simple reason: it only provides what is predictable, what fits pre-established patterns in its programming and/or conforms to clichés of mathematical logic — all expressed in its characteristic polished corporate jargon.

Can AI Think the New?

But can it create something genuinely new? Can it discover a new method, theory, vision, or offer a truly original perspective compared to what already exists? Can it produce something genuinely “intelligent” (by human standards)? No, not at all. Here, AI resembles a mirage in the desert: with every product update, every new version, we are presented with grand promises, spectacular capabilities, and impressive demos created by the industry’s “smart guys.” Yet when we try to use them ourselves, they are nowhere to be found. In reality, they simply do not exist. In other words, the trailer is brilliant, but the film is terrible.

Huge Costs, Uncertain Revenues

AI companies spend enormous sums developing and delivering AI systems, yet the revenues generated remain modest and cover only a small fraction of costs. Worse still: despite its much-advertised “intelligence,” AI has not yet invented any method of covering even its own operating costs, let alone generating sustainable profit. As a result, companies rely on the same marketing and commercial strategies taught in the first year of business school (if not earlier). When that is not enough, they resort to well-known commercial tactics — some bordering on deception — such as “shrinkflation”: reducing value while keeping prices the same. AI companies go even further. For example, Anthropic recently announced, with great fanfare, that it had increased its token capacity while simultaneously lowering prices. Users who took this at face value later discovered they were paying significantly more for less usage.

The Gap Between Lab and Reality

Despite all this, AI companies and so-called “experts” (whether “smart guys” or “useful idiots” of the sector) continue to claim that AI is “intelligent” and becoming even more so, according to scientific and mathematical benchmarks. Meanwhile, users tend to classify AI more as “the village idiot” — judging performance by human standards. For them, the initials AI are better interpreted as “Artificial Idiot” rather than “Artificial Intelligence.”

No Chance of Improvement

And things are unlikely to improve — quite the opposite. Today’s AI is already known to be a form of “mathematical-logical intelligence,” inherently limited compared to human cognition (a topic discussed in detail in Artificial Intelligence: Reality or Illusion). On top of that, there is the steadily declining quality of training data. In the early stages, high-quality sources were used (literature, scientific works, etc.). Today, this is no longer the case. Why? Because most valuable content — whether written, filmed, or otherwise produced — is protected by copyright, and companies would need to pay to use it. Something they are either unwilling or, more likely, unable to afford. As a result, many such works are used without permission, leading to numerous lawsuits, some of which have already gone against AI companies. For instance, Anthropic reportedly had to pay nearly $2 billion in damages for training its AI on illegally downloaded books. Additional cases are still ongoing, including lawsuits brought by major media organisations such as The New York Times and CNN (against Perplexity for unauthorised use of journalistic content).

The Snake Eating Its Own Tail

Under these conditions, AI companies increasingly turn to publicly available, free content — including social media. In fact, Elon Musk’s acquisition of Twitter (now X) was largely motivated by the desire to secure unrestricted access to platform data for Grok, his AI system; the political use of “X” is merely a bonus. However, this comes at the cost of training data quality, since most social media content is of mediocre or even negligible value. Worse still, an increasing share of this content is now generated by AI itself — raising growing concerns that future AI systems will be trained primarily on AI-generated data. A serpent eating its own tail. This creates a striking paradox: on one hand, we are encouraged to consume more AI; on the other, companies urgently ask for more “human” content to prevent AI systems from “hallucinating” too much.

AI “Hallucinations”

The fact is that AI — the great digital oracle claimed to know everything about everything — routinely invents bibliographies with the same ease as a school student invents excuses for not doing homework. It also produces quotations from non-existent works, fabricates sources, and mixes fact with fiction in the same breath. In doing so, AI becomes a professional generator of scientific fiction. Even more poetic are those links that lead to real articles unrelated to the topic — essentially saying: “Yes, it exists… but it’s something else entirely. Good luck finding the connection.” And yet, it cannot really be blamed: AI appears convincing precisely because it presents fiction as fact — and apparently “verified” fact at that. A talent humans recognise under a simpler name: lying. Which raises the question: has AI finally succeeded in copying the human mind?

The Illusion of Coherence

AI “hallucinates” due to a structural flaw: it is designed to respond in the most statistically appropriate way according to mathematical logic. It selects what appears most likely or most frequently represented in its dataset relative to the prompt. When it cannot find an exact match, it chooses what seems most plausible — even if unrelated to the subject. Moreover, its dataset contains numerous incorrect, misleading, or context-dependent pieces of information. Humans can usually navigate such ambiguity using experience, intuition, emotion, and critical reasoning. AI cannot. It does not judge. It does not verify. It assumes. If a quotation is attributed to a work, AI will not check whether the work exists. It simply accepts it as true. Thus it fabricates, invents, and “hallucinates.” In short, it cannot distinguish truth from falsehood when this distinction is not explicitly encoded. Nor can it handle nuance or subtlety. Therefore, it is — in this sense — “idiotic.”

When AI Becomes Too Convincing

Its architecture is, however, flawless in one respect: it is designed to appear authoritative. It creates the impression of certainty, precision, and truth. As a result, it is not difficult to imagine companies holding crisis meetings after AI recommends purchasing a product that does not exist. The plan is implemented in detail — budget, logistics, training — only for reality to eventually catch up. The inevitable executive question follows: “What do we do when AI is this spectacularly wrong?” The short (and cynical) answer: publish a viral AI-written article, only to later discover that half the bibliography is pure fiction.

Should We Abandon AI?

It is worth noting that AI systems are becoming increasingly “clever” in their own way. For instance, Gemini explicitly warns users that it “may make mistakes, including about people.” In other words: “If you trust me, you do so at your own risk.” From here, we can either embrace this technology and continue using it or reject it entirely and discard it. Our position is clear: it should be used. A lot. Intensively.

Use AI Widely, But Wisely

AI is not true intelligence, and it will not turn us into a hybrid genius — Einstein, Da Vinci, and Shakespeare combined. It is, however, a powerful and efficient tool. In essence, it is a hammer — a 2.0 hammer operating in the digital world. Used poorly, it can reduce intellectual effort, waste resources, damage reputations, and more. Used wisely, it can help produce complex work faster, at lower cost, and with less effort. But this requires more than surface-level familiarity or social media tutorials. It requires study, analysis, and experimentation to identify which AI tools best fit each task. Often, multiple systems must be combined, each specialised for a different component of the work. And above all, they must be used intelligently.

Case Study: “Coding”

According to AI companies and the so-called “smart guys,” AI-assisted coding is so simple that anyone can do it. In their view, AI does everything: you just tell it what you want and wait calmly for the desired software to be produced. This is a lie. Letting AI handle the entire process on its own can have disastrous consequences — from wiping a company’s database to producing bug-ridden, unusable code capable of breaking otherwise perfectly functional systems. This is why one AI expert advises: “AI — let’s say Claude.md — writes code. The human designs the system in which AI writes code.” In other words, AI writes lines of code, but under human supervision and control, while we handle the rest of the work. And coding itself represents only about one third of programming. The rest includes planning, writing technical specifications — tasks that only humans can properly perform (even if assisted by tools, including AI). In conclusion, AI is not a simple “artificial idiot,” but rather a “functional artificial idiot”: capable of excellent results, provided it is used intelligently.

How “Intelligent” Must We Be?

If we simply ask AI to perform a task, it will not necessarily know how to generate quality output. Worse, even for simple requests it may deliver nonsense. To avoid this, AI must be carefully “prepared” upstream. The words used to describe the task must be as simple and unambiguous as possible — no extra terms, no ambiguity, nothing that could “lose” it along the way. The request must be extremely clear and simple, with precise and complete instructions, expressed in straightforward language and conveying a single message. AI does not understand nuance, subtlety, irony, or figurative language; confronted with such elements, it is about as “intelligent” as a pile of scrap metal. Finally, the result must be carefully checked to ensure it matches the original intent and that no “hallucination” has corrupted the output. In short, we must be more “intelligent” than AI — meaning we must be capable of performing the task at least as well, if not better, than it can. For complex tasks involving multiple tools, including AI systems, we must know how to select the appropriate tools, combine them effectively, manage their interaction, and verify the final output. This is the minimum required for producing quality work with AI — and it becomes even more complex in the case of companies.

The AI Ecosystem

Today, before our eyes, a vast AI ecosystem is developing at breakneck speed. At its centre are AI companies (those that create and develop the technology), surrounded by a growing number of actors with vested interests — the so-called “stakeholders.” Among them are influencers — the “smart guys” around the AI industry — who claim that it is now impossible to live without using AI 24/7. There are also hardware companies such as Nvidia, which produces the chips on which AI runs; investors and financial institutions such as banks and investment funds; and construction companies building the enormous data centres that host the servers running AI systems. All of them are increasingly pushing us — today, and even more tomorrow — to adopt AI, both as individuals and especially as companies or corporations, because this is where their profits will come from, if such profits materialise at all.

“Loves Me? Loves Me Not?”

It all began as a murmur in 2025: companies must adopt AI quickly and at scale. If they do not, they risk losing productivity, market share, profits — and eventually going bankrupt. Soon, this murmur will turn into a deafening roar that, by comparison, makes Metallica sound like a trio of retired pensioners playing softly in a corner. In our view, this is precisely what should not be done. Following this path — strongly recommended, even imposed, by the increasingly loud voices defending the interests of AI companies (and by extension, the entire AI ecosystem) — amounts to guaranteed economic self-destruction. AI adoption should not be rushed, nor implemented under pressure or on the assumption that “everyone is doing it” (a narrative mainly promoted within the AI ecosystem and by “useful idiots”). Companies must adopt AI — but only in a structured, intelligent way.

How Artificial Intelligence Should Be Implemented Properly

AI implementation must be tailored to the specific needs of each company, not based on recommendations from AI ecosystem actors. At a minimum, several rules must be followed. First, companies must precisely identify the processes and tasks that AI can perform more efficiently, more quickly, and at lower cost than humans. Second, they must choose the most suitable AI system in terms of performance, capability, and cost. Third, personnel must be trained to use AI correctly, as discussed above. There must also be strict rules governing AI interactions with the outside world — especially with its creators — in order to prevent company data and know-how from being “siphoned off” to train future AI models or made accessible to other users of the same system, or even sold to third parties such as data brokers. Additionally, employees must be prohibited from using unauthorised AI systems for professional purposes. A clear separation must be maintained between AI used privately and AI used professionally, in order to prevent incidents such as data leaks or unintended cross-system interactions.

The Backup Plan

A backup plan is always necessary. Today’s AI is far from perfect and may fail unexpectedly. Moreover, AI companies are not exactly paragons of stability and may dramatically increase prices overnight, blowing up corporate budgets. Anyone relying significantly on AI must therefore have an alternative ready: either switching to another AI system or reverting to non-AI workflows. In our view, the best solution is to develop one’s own AI system — either independently (if possible) or in partnership with other companies to reduce costs and increase the chances of success. Such a customised AI would likely be less “idiotic” than commercial systems and would probably also be less vulnerable to the bursting of the so-called “AI bubble.”

What Is the “AI Bubble”?

An economic bubble is a phenomenon in which the price of an asset far exceeds the profits it is expected to generate in the future. In other words, there is a growing gap between the real value of an asset and its market price. In the early 2000s, we saw this phenomenon with internet-based services — the famous “dot-com bubble” — which ultimately led to the collapse of companies that once seemed eternal, such as Enron and WorldCom. Today, we are facing an “AI bubble,” in which AI is the overvalued asset, and companies such as OpenAI and Anthropic are among those best positioned to disappear if the bubble bursts.

Why?

As of June 2026, AI is a gigantic money-consuming machine that generates very little profit. There are only extravagant promises of a “bright, shining green tomorrow” (green being the colour of money). This “bright future” will only materialise if AI is adopted on a massive scale and if users — especially companies — agree to pay exorbitant costs. In our view, this is a chimera. If this does not happen, the “AI bubble” is likely to burst, leading to the collapse of AI companies and those dependent on them, as well as massive losses across the wider ecosystem.

Will the Bubble Burst?

Very likely yes — because AI has so far demonstrated a clear inability to generate meaningful profit for its creators. This situation will persist until the miracle occurs whereby AI evolves from “the village idiot” into a quantum physics doctorate holder — which is, apparently, only a matter of time and patience. We are still living in the era of inflated promises, where every algorithm is presented as a revelation and every chatbot as a genius. Each new version of AI is announced as bringing “the end of humanity (or medicine, work, cybersecurity, the arts) as we know it.” But reality is far less spectacular. AI performance remains modest and far below these exaggerated claims. And someone, eventually, has to pay for promises that never materialise.

Winners and Losers

AI companies, led by OpenAI and Anthropic, are still holding on, generating noise and launching increasingly bombastic announcements showcasing AI capabilities worthy of science fiction. Their survival depends on how long the “suckers” — investment funds, banks, other companies, and states — continue to finance them. When funding stops, the “bubble” will burst, and AI companies — with OpenAI and Anthropic at the forefront — will go bankrupt. These will be the losers. The winners will be the AIs belonging to companies with “strategic depth”: Microsoft, Alphabet, or Elon Musk’s SpaceX. These are the players that will endure and shape the post-bubble landscape. And they will naturally harvest the “spoils”: databases, software, engineers, and customers from the defeated firms.

A Slow Evolution Toward Real Utility

And yet, AI technology will continue to advance step by step, through accumulation, much like the internet, which did not appear overnight but developed gradually over time. We believe AI will evolve, improve, and eventually reach a point where a solid and realistic economic model emerges. At that stage, it will become an even more useful tool for humans. In the meantime, however, we humans will need to learn how to use AI responsibly, in compliance with legal and moral frameworks. At the same time, AI companies will need to significantly strengthen the internal safeguards of their systems, with the goal of preventing illegal or unethical use. TO BE CONTINUED.