Thomas Ptacek: A LLM for Correction, Not Writing

Developer Thomas Ptacek advocates for a limited use of language models focused solely on correction and error detection, without allowing them to write a single line. He publishes a prompt and showcases his own proofreading tool to facilitate this approach. His red line: never adopt a phrasing suggested by the AI, in the name of quality and discipline.

He shares a personal tool and a proofreading prompt

Thomas Ptacek presents a screenshot of his personal text correction tool using an LLM. He provides a prompt to help kickstart the creation of a similar tool, as well as a proofreading prompt to frame this usage.

Authorized uses: fact-checking, spelling, grammar, thesaurus

In his approach, Thomas Ptacek restricts LLMs to fact-checking, spelling and grammar correction, and occasional use as a thesaurus. He clarifies that he does not allow LLMs to write the content of his blog and considers them as proofreaders rather than writing assistants.

A strict rule against any phrasing suggested by the AI

Thomas Ptacek states a central rule: never use a single word or phrase suggested by an LLM, and apply this prohibition strictly. He believes this rule is wise, that texts generated by LLMs have a particular "smell," and that this discipline helps maintain control over the quality of the text.