LLMs solve the biggest problem with language textbooks (til.andrew-quinn.me)

Large language models (LLMs) have plugged the biggest practical gap in textbook-driven language learning: instant, reliable feedback. Where learners historically had to wait for a teacher or fluent speaker to check exercises, learners can now work through a textbook and immediately verify answers, get corrections, and request explanations or alternative examples from an LLM. That changes the utility of carefully authored secondary‑level textbooks, making them far more viable for self-study and shifting the meta‑competition among learning materials toward those with clearer, progressively scaffolded exercises that pair well with automated feedback. Technically, the benefit depends on model coverage for a language: high‑resource languages (English, Spanish, Arabic, Norwegian, etc.) get much better corrections and explanations because LLMs were trained on abundant data, while rare or endangered languages remain better served by human tutors. Imperfect outputs don’t negate value—models can still point out errors, paraphrase prompts, generate plausible practice dialogs, and offer targeted grammar notes even if they occasionally “whiff” fine points (as has been observed with complex Finnish morphology). For authors and learners alike, the implication is practical: combine traditional textbooks with LLM-driven checks and iterative prompting to accelerate acquisition, reserving human experts for nuanced feedback and low‑resource cases.