I wrote an AI textbook — how long until AI can do it better?
AI models still can’t write solid long‑form non‑fiction, so authors must keep the human voice for insight and structure.
Why Now
The author finished a 2026 textbook using LLMs, noting that recent GPT‑4.5 and Claude models haven’t improved long‑form writing much.
What Happened
LLMs excel at spotting typos and editing short passages but struggle to organize chapters, leading to muddled structure and conceptual errors. The author used Claude for editing and GPT‑5.5 Pro for deep typo detection, but only <1% of AI‑generated sentences made the final book. Current models lack training data and inference‑time scaling for high‑voice, long‑form writing.
Why It Matters
Because models can’t yet synthesize coherent, insightful non‑fiction, scientific progress that relies on autonomous knowledge synthesis will remain limited. Human guidance will stay essential for producing reliable, structured content and for catching subtle errors that AI misses.
The Limitation
The analysis is based on a single author’s experience and may not generalize to all writing styles or future model releases.
What You Can Do
Try using an LLM as a focused editing assistant—feed it specific sections and let it flag typos or suggest phrasing, but keep the overall structure and voice under human control.
Source
Read original sourceWhy we picked this
Reflection on AI’s writing ability is meaningful AI news about capabilities.