The reality is that we are facing a dead end

There is a race among major AI companies to build a AGI or artificial general intelligence, that which surpasses humans in all areas of knowledge. Sam Altman has been hyping up about AGI for a long time, and he is not the only one. Mark Zuckerberg has spent a million on a team to create it, Dario Amodei believes it will arrive very soon and Elon Musk says Grok 5 could achieve AGI. What if it’s all a big lie?

Language and intelligence. It’s not the same. OpenAI, Meta, Anthropic… all these companies have something in common: their path to AGI is LLM or great language models. They count in The Verge The nuance, and it is not just any nuance, is that language and intelligence are two very different things. Decades of research have shown that language is a tool of communication, not thought. In other words, mastering the language is not equivalent to more intelligence, in the same way that not mastering it does not mean ceasing to be intelligent.

Language models. ChatGPT, Claude, Gemini… are tools composed of hundreds and even billions of parameters, trained on enormous textual corpora. Their technical complexity is undeniable, but they are still systems that They predict the next word from statistical correlations. Its core is language, not ideas or abstract thought in the human sense.

Achieve AGI. Getting to AGI with a language model is a dead end. I said it recently Yann LeCunconsidered one of the godfathers of modern AI and, until recently, head of AI at Meta. According to LeCun, the path to achieving AGI is not the LLM, but the LWM or world models. These models they learn from the environment and they can imagine scenarios, like humans do.

He hype. If language models are not the way, why do AI companies keep saying they are on the verge of achieving AGI? Because they need it. Their premise is that with more computing power (more chips and more data centers), their AIs will become smarter, so fueling the hype is their way of justifying that absurd amounts continue to be invested to scale AI.

Deceleration. At first the evolutionary leaps in AI chatbots were palpable; the first versions of ChatGPT they blew our minds. The reality right now is that Generative AI has entered a stage of continuity or deceleration. There are improvements, but they are no longer as notable or revolutionary. The solution is to generate expectation: with AI agents and, of course, with the AGI.

Does this mean that AGI will never be reached? Not necessarily, but it will take more than language models and above all time. According to Andrej Karpathy, co-founder of OpenAI, AGI will take for at least another decade.

Image | Meo, Pexels (edited)

In Xataka | Someone has taken a look at the dotcom bubble and compared their data to AI. And it’s not optimistic