videlalvaro

Once I got API access to GPT-3 the idea for what to build was very simple, make the algorithm write a XVII century Spanish novel. Considering the modern novel appeared about that time, I wonder what this modern algorithm could produce.

My intent was to know if an algorithm like GPT-3 would be able to know Spanish, recover the Catholic faith, fight against the Moors or the Turk, forget the history of Europe between the years 1602 and 2020, be Miguel de Cervantes. That of course what too easy. Not even a challenge for an algorithm that possesses today’s modern databases.

A very different challenge would have been to be a XXI century algorithm, knowing everything there is to know about Spanish and novels: the Spanish standardization of the XVIII century, the Latin American grammar of Andrés Bello, the Realismo Mágico of García Márquez and Rulfo, the poetry of Mistral, Ibarbourou, Walsh. To know all that and still write a XVII century novel? Well that would set GPT-3 apart.

As soon as I typed “write a XVII century Spanish novel” and pressed enter, I was surprised by the results:

truth, whose mother is history, rival of time, depository of deeds, witness of the past, exemplar and adviser to the present, and the future’s counselor

Which compared to Cervantes Don Quixote is fantastical:

truth, whose mother is history, rival of time, depository of deeds, witness of the past, exemplar and adviser to the present, and the future’s counselor

The contrast in style is also vivid. The archaic style of GPT-3—quite foreign, after all—suffers from a certain affectation. Not so that of his forerunner, who handles with ease the current Spanish of his time.

GPT-3 has (perhaps unwittingly) enriched the slow and rudimentary art of reading by means of a new technique—the technique of deliberate anachronism and fallacious attribution. That technique requiring infinite patience and concentration, is but a blink of an eye for an algorithm like GPT-3.