A Safe Way To Let Generative AI Edit Your Document

Can you let generative AI edit your Word document? Sure, if Track Changes is on. But there’s a cost.

An analysis need

I had a multi page Microsoft Word document, originally written by a client and subsequently edited by me in two separate sessions. This version, 1.2, desperately needed a sanity check—analysis, if you will.

  • First, to see if it made sense.
  • Second, to see if it properly addressed two distinct sets of hungry people (target audiences): the business audience interested in the brief introduction, and the technical audience interested in the main body of the document.

I wanted to use a tool to perform the work, but I didn’t want the tool to just make the changes on its own. All I need is one buried hallucination and my client’s credibility would be ruined.

Google Gemini.

Suggesting the changes

So I told the tool that I wanted to create a version 1.3, and to compile a list of suggested changes for my review, describing the two sets of hungry people for portions of the document, and capping the list of suggested changes at 40.

It created a list of 38 suggested changes, within my self imposed limit.

Google Gemini.

But before I could review the list, the tool asked me a question (paraphrased):

“Do you want me to apply the changes and create a marked up version 1.3?”

Making the changes

Those words “marked up” gave me joy. If you haven’t worked with Microsoft Word, it has a “Track Changes” feature to let you know every change someone (or something) made to a document. You can either accept them or reject them.

Google Gemini.

For me it was the best of both worlds. The tool could automatically make the changes without forcing me to make them manually. And I could review its work.

So I let the tool make the 38 changes, and I waited. And waited. (And spent an inordinate amount of tokens.)

But the tool delivered my 1.3 document, all marked up.

Google Gemini.

Reviewing the changes

I accepted most of the changes, with one major exception.

You see, in my first round of manual edits I had changed most of the section numbers, failing to realize that the text included hard-coded references to the old section numbers.

  • The tool fixed these so that the hard coded references were now directed to the new section numbers. That was fine.
  • But then the tool inserted hard coded references to the page numbers. That was NOT fine, since this is a living document that may expand or contract, changing the pagination.

It cost you

But all in all the markup looked good, was accurate, and saved me and my client time. But it did have a cost.

Google Gemini.

Now I just have to decide when to spend tokens and let the tool create a marked up version.

By the way, here’s a video summary.

Let Me Do It. Google Lyria. Public Domain.

So what?

Of course, my view of generative AI use falls within these two extremes.

  • If you even think of using generative AI in any way, you will never consult for us again.
  • Your performance is measured by the number of tokens you consume.

Yet in reality, my use or non-use of generative AI is extremely inconsequential. The much more important factor is my expertise in identity, biometrics, and technology.

If you want to harness that expertise for your firm, talk to me.

https://bredemarket.com/mark/

Leave a Comment