Cyborg-Fagia
One of the simplest ways to imagine how large language models (LLMs) work is to think of them as a digestive system: an LLM receives our words, deciphers them, processes them, and returns an output shaped by the text we previously provided.
That metaphor brought to mind the premises Oswald de Andrade articulated in the Anthropophagist Manifesto, published in 1928. Oswald, a poet, essayist, playwright, novelist, journalist, and theatre critic, was a key intellectual and theorist of Brazilian Modernism, and one of the organizers of the 1922 Week of Modern Art. The manifesto first appeared in the Revista de Antropofagia and became one of the most cited texts in Brazil’s twentieth-century cultural imaginary.
The text is a defiant declaration against an artistic elite positioned at the heart of Brazil’s cultural and artistic establishment, whose aesthetic and stylistic values were deeply grounded in the idealization of European canons of beauty. Its call was to rebel against that external aesthetic authority through a cannibal practice.
The manifesto’s most emblematic line is a reconfiguration, a recycling, of Shakespeare’s classic “to be or not to be.” De Andrade takes it, processes it, deciphers it, and spits it back out transformed into a Brazilian linguistic icon: “Tupi or not Tupi, that is the question.”
In that gesture, he transforms the European-Anglo “to be” into “Tupi”: the situated subject of the Amazonian Brazil, the Indigenous being. In this way, the Anthropophagist Manifesto, more than an aesthetic manifesto, becomes a decolonial metabolic machine that offers an alternative to merely rejecting the European canon. Instead, it chooses to consume it, metabolize it, and transform it in favor of creating a new symbol: an object of local Brazilian expression.
The comparison I propose between De Andrade’s cannibal practice and the functioning of language models carries a risk: it can depoliticize the cannibal gesture, anthropophagy itself. Language models have no bodily experience and are not subjected to the needs and conditions that define our human nature; therefore, the analogy could slip into a kind of bulimia, where automatic repetition of consumption turns the gesture into an empty repetition, stripped of the most important action in anthropophagic practice: symbolic appropriation.
But this resolves the moment we stop thinking of language models as a complete digestive system and instead place them within a symbolic chain of meaning production, as one element that can help us carry out our anthropophagic practice as situated subjects.
And that subject is not just any user.
Here, human existence itself, lived experience, rises as a political element. Because it is not the LLM that defines the meaning of what it produces: it is we, as situated users, aware of our perspectives, experiences, preferences, opinions, and reasoning, who decide the validity of the outputs it gives us.
Consequently, that decision is not neutral. This is what makes human existence political in itself, because a situated user is neither a passive consumer nor an admirer of the model’s statistics. Rather, it is a user who enters the platform to stress the LLM, to demand positionality, to demand that their own imaginary directs the metabolization to be carried out. This user does not come to adapt to the language model; they demand the calibration required by their own experience.
Therefore, it is not enough to use these technologies passively. What is required is to use them from subalternity, so that the center is not allowed to keep reproducing itself.
To use artificial intelligence as a situated user is a gesture that bends the anglocolonial bias of corporate models and forces them to work with us as collaborators in an anthropophagic practice.