Harnessing AI Tools for Writing
I agree with the “fact” skepticism but realize that frontier models are equipped with tools that enable them to search the internet and summarize the results with attribution. So, it doesn’t have to rely on training data.
For example, with claude research you can pose a premise (I’ve learned to use a prompt that asks for evidence of validation and refutation at equal weight). And you get back a summary with validation links to the papers cited.
The human skill is how to use that information. To validate data, challenge bias, and engage cognitively with the information.
I download (purchase) the papers and read the relevant sections. I ensure I fully understand the findings I use as support for my writing. Whether I summarize the finding myself or use the agent for a first draft, I actively edit multiple drafts using the agent as a sounding board for devils advocate and to catch mistakes. All in addition to traditional grammar/spelling check and as much as practical, peer or editorial feedback.
I make sure my work is my voice and my writing and I honor whatever guidelines are imposed on the writing. I disclose how I’ve used llm assistance.
You didn’t say this but outright prohibiting the use of generative tools is a lost opportunity. Preventing a person from using a tool and denying that it is helpful when at an initial glance it can spit out the whole paper and businesses are boasting how much they rely upon AI will create a dissonance and incentive to cheat.
Better to teach people how to use these tools. How they aren’t magic. They do a ton of mediocre work. And if not actively overseen by the human will foster negligence. But they are powerful tools for reducing what many find to be drudgery in the writing process. Gathering data that needs to be vetted. Suggestions that need to be evaluated. Draft language that needs to be reviewed and redrafted.
In disclosure I used an llm, not to write this but to give me a devils advocate response which influenced a second next draft.
