Using ChatGPT as a technical writing assistant

chatgpttechnical-writingai-assisted-writingprompt-engineeringworkflowthoughtworks

Abstraction: ChatGPT as iterative technical writing draft tool

Key points:

Mike Mason (Thoughtworks) used ChatGPT to draft Technology Radar "blips," a Macro Trends article, and business-audience rewrites; avoided all client/proprietary content for IP reasons.
Strengths: instant recall, good first drafts, useful suggested sections (it proposed "goals and expectations" for this very article).
Weaknesses: accuracy/depth gaps and info-cutoff, overly "bubbly"/optimistic tone (words like "novel," "unique" needed removal), context limits (GPT-3.5 vs larger GPT-4), and "anchoring" where it repeats content instead of expanding.
An ML expert reviewer said a Giskard blip read as written by a junior with little ML-workflow understanding — output needs skilled expert review.
Recommended strategies: reusable context-setting prompt, work iteratively in small chunks, treat it as a peer (models respond at your apparent level), start a new chat to escape anchoring.
Conclusion: didn't save net time but enabled steady progress; first-draft/suggestion use of LLMs works broadly, and a human expert remains necessary.

Connections: Chatgpt · Thoughtworks · AI Assisted Writing · Prompt Engineering · Human In The Loop