Stargate's $500B AI Bet Amid AGI Skepticism

The promised AGI and the end of work nobody signs up for

The announcement of the Stargate project, a $500 billion investment in artificial intelligence infrastructure in the United States, has peine a debate that had been simmering for months: if general artificial intelligence truly arrives, what happens to jobs, the economy, and the very idea that we are special? The official promise is ambitious: build AGI and cure diseases at an unprecedented pace. The reaction, in digital forums where this is discussed without filter, ranges from apocalyptic enthusiasm to ruthless mockery. In the middle lie uncomfortable questions that no one fully answers.

What is AGI and why its antiestéticasibility is debated

General artificial intelligence is defined, in the generated conversation, as an AI capable of performing any intellectual task a human can do. Popular imagination turns it into an almost omniscient god one can converse with. The first objection is technical: an AI, however advanced, does not really understand what is asked; it only statistically predicts the next word. The second is more philosophical: to have AGI, one would need to emulate the human brain, based on carbon and with its chemical capacity to form complex and stable structures, something argued to be very difficult to replicate in terms of reflective consciousness.

Counterarguments arrive quickly with a repeated analogy: they said the same about birds and flight a hundred years ago. It is not about emulating the brain's internal functioning, but simulating its external result. Early planes did not flap wings; artificial neural networks have nothing to do with biological neurons, they are just interconnected and pass information from node to node. From outside in, not inside out. The future, declares one of the most vehement voices, is data.

The consciousness debate: do humans reason?

One of the discussion's most interesting twists shifts focus from the machine to us. Language models do not reason because they do not understand, it is said. But do humans reason? The answer thrown out is uncomfortable: 98% do not; they use common sense and intuition, making decisions based on patterns of prior experience. In that area, models already surpass us. Many discoveries and innovations may simply be the fruit of post-hoc rationalization: first we intuit, then we reason and find proof. Recent models with chain-of-thought and mixture-of-experts are basically that: rationalization.

Some take the argument to the extreme, claiming no person on Earth possesses even 1% of the knowledge accumulated by a language model, and that the day we discover we also operate via statistical prediction will shock many. Coherence, another adds, is inevitable.

The economic scenario: four groups and a basic income

If AGI arrives, how is the pie divided? The most repeated prediction sketches a capitalism with handouts and four well-differentiated groups. First, the super-rich. Second, minor rentiers: people with stocks or housing who, with increased productivity, will live well without working. Third, civil servants, who will pretend to work but get paid quite well, as now. Fourth, specialized irreplaceable workers. And fifth, recipients of a universal basic income living in cramped apartments with food and entertainment, doing nothing, so they don't bother anyone. Basically, what exists now but removing most of the population from the labor market.

The problem with that scenario, it is objected, is that hungry people tend to cause trouble. Therefore, the speculated plan is not Elysium, but keeping the masses entertained: universal basic income, non-reproductive sens, porn, cheap entertainment, and legal drugs. A massive castration plan where, in two generations, the oligarchy inherits a clean Earth devoid of mediocrity. World population reduced to a tenth. The question, it is admitted, is whether that is sustainable or if the invention explodes before then.

The data problem: hitting a ceiling

Against the optimism of big numbers appears the most grounded argument: models are not in diapers; they are hitting a ceiling. Even Elon Musk acknowledges there are no quality data left to train them. In that scenario, eliminating 80% of potential content creators for these AIs sounds like a self-fulfilling and rather stupid prophecy. Charlatans, it is ironized, turn smoke into gold with astonishing ease, and Trump has just promised them a fortune.

This is compounded by suspicion regarding data bias. If you ask an AI about genocide in the Spanish conquest of America, it will say yes, cite paragraphs, and even links. If you ask about Anglo-Saxon colonization, it will respond that history is complex and diverse, going off on a tangent. Whose data is it? That is the question.

Spanish AI: ALIA and the LLM that fails to take off

In Spain, moves have also been made. The Government presented ALIA, an open-source, publicly funded Spanish-language language model. Reception has been, let's say, lukewarm. Early tests speak of inconsistencies, poor Spanish writing, and, to no one's surprise, it is a fine-tune of Llama, Meta's model. Indra allegedly put in three interns and pocketed some cash, critics joke, without proof but also without doubt. The repeated diagnosis: in Spain, AI is trained with PSOE dogmas, phrases from Charo and Yoli, inclusive language, and Newtral verifications. Then it will come out zascandil, and blame will be placed on US rich.

Meanwhile, the fundamental question remains unanswered. AGI will be here soon, they say. Or not. The exact point where analysis stalls is the same where it started: no one knows if the bottleneck is technical, economic, or simply human.