I have written four pieces in recent weeks. I did not plan them as a set, but rereading them I see they say the same thing from different angles. This is the thread that connects them.
First thesis
The problem is not capability — it is abstention.
Language models can speak about almost anything. That is their virtue and their trap. When they don't know, they don't stop: they complete. They produce an answer with the texture of truth that isn't true. The failure is not technical — it is the absence of a mechanism that says: I don't have sufficient basis here — I stop. Wittgenstein closed the Tractatus with seven words about silence. Building that silence into an AI system is probably the hardest problem in the field — and the least profitable.
Read → Silence as Technical Capacity
Second thesis
The first generation solved access — the second faces judgment.
The digital humanities did immense work: digitizing, indexing, making searchable what once required physical presence. Thanks to that effort, the archive is now available. But that very availability is what makes the problem critical: when the machine can speak fluently about any corpus, the question is no longer can it answer? but should it? The leap is not technical. It is epistemological.
Read → The Question Has Changed
Third thesis
Models master correctness — but what matters is clairvoyance.
Bolaño distinguished between placing the right words in the right place — which is correctness — and clairvoyance, which is something else. Models add by default. They append. They complete. But they don't know how to subtract with intention. They don't distinguish between something missing by oversight and something missing on purpose. For them, an absence is always a gap to fill, never a decision. Restraint — knowing what to leave out — is still ours.
Read → Bolaño, Clairvoyance, and Artificial Intelligence
Fourth thesis
If transmission is automated, the question is what the university does with the time it has left.
AI tutors already do the work of explaining, exemplifying, personalizing pace. The university cannot compete with that — nor should it try. What it can do is use the freed time to cultivate the one thing the machine does not have: judgment. Training researchers, not content consumers. The bottleneck is no longer explaining. It is thinking.
Read → From MOOC to AI Tutor
All four theses converge on a single point. The problem with artificial intelligence in knowledge domains is not a problem of computational power or model size. It is a problem of judgment: knowing when to stay silent, distinguishing between evidence and fluency, subtracting instead of adding, and deciding what to do with the time the machine frees.
Ateneo — my editorial verification instrument for humanities corpora — is the attempt to take that problem seriously. Not as opinion — as architecture.