DocWriter team describes a multi-agent approach to learning a user's writing style
A commenter argues that a “neither is good” option matters: otherwise, the less awkward passage could be recorded as a style the user actually wants.
TLDR
DocWriter’s team says it built a pipeline of AI agents that analyzes word choice, grammar and how writing is organized, extracting patterns with supporting evidence. Users can confirm or reject each pattern by comparing two passages side by side. In the August 22 update, the team said performance in its next user studies remained an open question. A commenter highlights the importance of “neither is good”: without that choice, a preference dataset can mistake the less awkward sentence for a genuine style preference.
Combined views
1.1K
2 Sources, first seen 18d ago