He learned Portuguese by being wrong

Dreaming about Germany resurfaces his German, and the comparison between his two language educations is a controlled experiment with himself as both subjects.

Experiment one: years of formal German, Chía academy then five university levels. Outcome: the declensions never clicked (“me las memoricé”), the courses got passed, and the enduring result was fear, “siempre me dio miedo hablar, por miedo de equivocarme, y eso es pésimo para aprender idiomas” ▸ 24:54. The group setting made every error public; the oral exams made errors count ▸ 24:04.

Experiment two: Portuguese with Julia, fluent in about six months. His own diagnosis of why: “yo creo que aprendí portugués rápido también por eso, porque yo la cagaba muchísimo, y Julia como que: no, listo, tú solamente di esto” ▸ 25:32. Corrections without verdicts. No “¿cómo va a cometer ese error?”, just the fixed phrase and onward. “Uno aprende sin el miedo a cagarla, y uno la caga mucho, pero uno aprende” ▸ 25:53.

la variable no era el método; era el precio del error →

Same brain, same aptitude, wildly different outcomes, and the only variable that moved is the price of a mistake. Where errors cost grades and face, the learner optimizes for silence, and silence is zero practice. Where errors cost nothing, the error rate goes up and so does the learning rate, because, as they note reaching for their “familia del futuro” ideal, the error is the information: “cuando uno tiene el error es cuando uno sabe: esto está bien y esto está mal” ▸ 26:04.

The design implication runs straight into Sanfanson, whose AI teacher grades every sentence: the product’s real spec isn’t the grading, it’s the tone of the grading. And the residue of experiment one is now half-usable anyway, German job descriptions read at 50% comprehension years later. Imagine what it would be if anyone had let him be wrong out loud…