Gary Marcus vs. Nature: Why rumors of AGI's arrival are premature

Gary Marcus and his colleagues, Walter Quattrociocchi and Valerio Capraro, have criticized a recent article in the prestigious journal Nature, which claimed the achievement of artificial general intelligence (AGI). The authors of the response publication insist that the impressive successes of large language models (LLM) on various benchmarks and even in solving complex mathematical problems do not constitute evidence of genuine intelligence. In their view, proponents of the idea that AGI exists are committing a fundamental error, confusing the ability to perform narrowly specialized tasks with the manifestation of truly general intelligence.

This article is a call for greater caution in the use of terminology and for more thorough, meaningful analysis of what we mean by the concept of "intelligence."

Recently, more and more people have been claiming that artificial general intelligence already exists. Perhaps the most recent and prominent such claim is contained in an article published in the journal Nature. These assertions are often fueled by impressive achievements in the field of large language models (LLM), whose results demonstrate high performance on various test datasets, fluent operation across diverse domains, and, in some cases, even correct solutions to open mathematical problems. These developments are often viewed as irrefutable proof that humanity has reached the threshold of artificial general intelligence.

However, as Marcus and his co-authors rightly note, such interpretations are based on a fundamental confusion of results from individual, often well-studied and standardized tasks with intelligence itself. Performing individual tasks, even if it demonstrates impressive results, cannot be considered sufficient evidence of the presence of general intelligence. In their article, the authors show that recent claims about achieving AGI are based on a conceptual error—confusing increasingly complex statistical approximation with intelligence itself. They also argue that recent claims (for example, published by Chen et al., 2026) about alleged success in creating AGI depend on redefining what the term "AI" has historically meant.

The main idea promoted by the authors is that modern LLMs, despite their striking capabilities, are essentially very sophisticated statistical machines. They are trained on vast arrays of data and learn to predict the next word or sequence of words based on probabilistic patterns. This allows them to generate text, answer questions, and even solve tasks that require a certain level of logic or knowledge.

However, according to Marcus and his colleagues, this does not mean that the model "understands" the task in a human sense or possesses the ability to transfer knowledge and skills to completely new, unforeseen situations—a key aspect of general intelligence. They argue that this resembles rather an advanced form of imitation or approximation than genuine thinking.

The consequences of such confusion can be quite significant. Premature claims about achieving AGI can lead to excessive optimism, incorrect allocation of resources, and, more importantly, underestimation of the real problems and challenges associated with creating genuine artificial intelligence. It can also lead us to be less critical of the capabilities and limitations of existing AI systems, assuming they possess a level of understanding that they actually do not. Furthermore, it can slow progress in research aimed at creating more reliable, interpretable, and truly intelligent systems.

In conclusion, Gary Marcus and his co-authors call on the scientific community and the general public to adopt a more sober and critical approach to evaluating achievements in the field of artificial intelligence. They emphasize that it is important to distinguish between the impressive statistical capabilities of LLMs and genuine general intelligence, which implies the ability to reason, learn, adapt, and understand across a broad spectrum of contexts. Until we reach such a level, claims about the advent of AGI should be regarded as premature and based on a misinterpretation of the data.