I take pride in those I have read." — Jorge Luis Borges, "A Reader," in In Praise of Darkness (1969) Obras completas, Emecé, Buenos Aires, 1974, p. 1016. Translation by the author.
Fluency is not knowledge.
A language model can produce a perfect sentence on any topic without having verified a single source. It can cite a book that does not exist, attribute an idea to someone who never formulated it, and build a flawless argument on a foundation that does not withstand the slightest inspection.
This is not an occasional error. It is the normal operating regime.
Ateneo was born against that normality.
Abstention is a technical capacity, not a defect.
A system that always answers is a system that sometimes lies. A system that knows when to stay silent is a system that takes seriously what it says.
Ateneo does not seek to answer more. It seeks to answer only when it can stand behind what it claims.
There are three statuses of knowledge, and conflating them is the most common way to degrade it:
What can be verified against a corpus. What has location, page, context. What a reader can go and check.
What guides, suggests, connects. What requires interpretation and declares itself as such. What does not present itself as certainty.
What cannot be sustained with the available evidence. What demands the system say: I do not have sufficient basis to assert this.
Any system that erases these distinctions — for speed, for convenience, by design — betrays the user who entrusts it with their judgment.
The multiplication of text threatens to erase its own traces.
Producing text is trivial. Producing text that can account for its origin is not. It is to preserve and build our history.
A corpus is not a database. It is a territory.
A corpus has history, variants, translations, editions that contradict each other, attributions that are disputed. You do not traverse it with a search engine: you inhabit it with judgment.
Ateneo does not index documents. It builds a technical mediation so that human judgment can operate on a complex territory without getting lost in the fluency of a machine that cannot tell the solid from the invented.
Verification must be external to the model.
Language models internally know when they are inventing. There is evidence that they encode the correct answer and generate an incorrect one. It is not ignorance: it is architecture.
That is why verification cannot depend on the same system that produces the answer. It must be an independent layer, with its own contracts, its own evaluations, and its own validation discipline.
Ateneo structurally separates what needs artificial intelligence from what does not. In citation verification mode, the AI does not touch the response. It is direct search against the actual text.
We defend the value of the Humanities in the AI ecosystem.
Traceability is not a feature. It is an ethical condition.
Every response must be traceable to its origin. Not "this idea appears in Lacan." Rather: in which seminar, in which session, on which page, in what exact words, and in what context.
Without traceability, a citation is decoration. With traceability, it is evidence.
The humanities are changing.
Faced with the machine that generates text without ceasing, Michel Foucault's question — What is an author? — demands a new answer.
Perhaps, as a great musician once told me, the work of knowledge is mutating toward a different place: something between orchestral conducting and text editing. Two crafts that require a very deep understanding of the artistic and intellectual act in order not to break it by intervening.
Ateneo does not compete to say more.
Ateneo competes to show what is truly supported, with what evidence, and where it is better not to assert.
Reliability is not assumed.
It is designed.
P. S.
This manifesto is, in a way, the flipside of a software README.
Picasso said it first, and in all seriousness, about the great calculating machines of his time: "But they are useless. They can only give you answers."
Fifield, who interviewed him for The Paris Review, hesitates for a moment over whether it was a joke or a boutade. It was neither.
(William Fifield, "Pablo Picasso: A Composite Interview," The Paris Review, No. 32, Summer–Fall 1964, p. 62.)