Gemini and I Wrote a Book: Introduction to Computational Linguistics (dubovik.eu)

A developer used Google’s Gemini to generate a full textbook—“Introduction to Computational Linguistics”—by building a small authoring pipeline: prompts defined in a YAML-like format, Python orchestration, human-readable caches, and LaTeX compilation. The run produced the book in 161 minutes across 351 queries at a cost of about €12, consuming ~745k input tokens, ~905k internal (“thought”) tokens, ~222k output text tokens and ~15.5k image tokens. The author needed custom serialization and LaTeX fixes, and deliberately separated prompt templates from code to keep the workflow modular and reproducible. The experiment highlights both the promise and limits of current LLMs: rapid, low-cost content generation democratizes projects that once required teams, but outputs are brittle—sensitive to prompt phrasing, weak at formal logic, poor at reliable citations (hallucinating references), and messy in orthography/markup. Practical takeaways for AI/ML practitioners: build explicit caching and validation layers, integrate retrieval (RAG) or persistent memory cautiously (legal risks noted), and combine symbolic/planning tools with LLMs for robust reasoning. Strategically, the work illustrates economies of scale in LLMs, potential market concentration, and the need for tooling and policy to ensure quality, traceability, and epistemic integrity when automating technical content.