Brad DeLong's June 4 essay links reading with language-model use, arguing that both depend on a person's effort to interpret words without direct access to the mind behind them.
The essay grew from a social media disagreement over whether imitation can teach someone a usable grammar. DeLong responds by comparing learning through a printed text with learning through a system that generates language. In DeLong’s comparison, neither reading a book nor talking to a language model puts the user in direct contact with a human mind behind the words. Meaning comes partly from the reader's work of interpreting what appears.
DeLong describes a more engaged form of reading than simply moving through a text. A reader may challenge an author's argument in their own mind and picture an answer, despite having no direct conversation with the writer. He points to Niccolò Machiavelli's 1513 letter to Francesco Vettori as an example of engaging with an absent author through books.
He extends this comparison to language models. DeLong says a model's words can prompt interpretation, but he does not treat the computation behind them as a human mind. He makes a related point about books: printed marks do not contain the writer, even when readers experience them as a voice. The analogy shifts attention toward the reader's role in making sense of language.
The June 4 essay followed a June 1 post about using a language model to work through a difficult sentence in Cicero's In Catilinam I. DeLong ends by asking what proficient use of language models involves and how people should learn it, while leaving both questions open.
