LLMs and Non-Fiction Writing: Stagnation and Opportunities

Carte mentaleMind map

The author, having used LLMs to write a technical manual, observes a surprising stagnation in their capabilities for long-form non-fiction writing. Unlike other fields such as coding or mathematics where progress is rapid, these models struggle to coherently organize and present established knowledge, often producing confused and disorganized content.

While LLMs excel at verification tasks (typos, minor errors) and targeted editing (responding to editor questions, suggesting rephrasing), they lack the ability to generate long, complex texts with a voice, point of view, or deep understanding. Their utility lies in assisting with repetitive tasks or transforming existing content, but they cannot yet replace human expertise for creating new knowledge or structuring complex arguments. The author believes that the best manuals will remain the work of humans for the next 2 to 5 years.