Some Thoughts about AI & Writing

AI is in everything, everywhere all at once.

There’s an interesting discussion happening online about writing. We obviously are seeing paid writers with multiple books suddenly releasing new ones they worked on using AI. And I don’t think people would have thought these accomplished writers would have used those services, but then again, they became lazy enough not only in writing but also in proofreading. They released books with the AI-prompt and the AI-generated preface to replies also present.

The trust is slowly eroding, because everyone feels and knows the grey area they are in. People don’t want to mention the extent of AI-assistance, as it will mostly backfire. Also, with AI as a buzzword is at the crossroads of what excites the next round of funders but also what curses the users, the rebranding of everything under the umbrella term of AI isn’t helpful.

At what point would word processor offering grammar correction would be renamed to AI assisted? We know Grammarly upped their offerings to be AI-assisted a while back, and this will be true for newer services too.

I think we all agree the part we need to hate about AI - is it generating anything from a prompt and it building something from nothing (okay the nothing here is the actual stolen data they’ve trained on) - and that’s why we need people to realise what part of AI is just a money-grab rebranding of features that was always available.

Humanizing the AI work, to pass AI test!

The other issue is tools that claim to help you check if the material is human-written or AI-drafted texts, and then humanize them, so they aren’t caught. This feels just the next layer of money grab scheme, because AI is generlaly bad at detecting such things, and someone willing to pay a genAI system to recreate the patterns so they appear more human generated content. My distrust stems from this: I’ve tried checking many of my old work from early 2000s and the checkers have been anywhere from 30% to 90% certain about parts or the whole thing were generated by AI. Maybe I was a young robot, with no clue.

Writing tools and version control

Honestly, my plan was to write about my preferred tool for writing, because another discussion that sparked concerns the need to keep drafts after iterative drafts of the process. I realise it works for most people since everyone tends to use Google Docs or Microsoft Word - both with their native version history. But I imagine there might be many who use the most basic tool with least amount of distraction, or just a notes app.

In my case, I love using LogSeq. It’s a note taking app, tht works by building an outline and then populating those parts. I pretty much work on one file and I rarely revisit a previous version in stories or blogs that I’ve discarded or reworked. While I must write a blog on all the advantages of LogSeq in my workflow, the most important one is that each file is saved as markdown and I can just copy the file to my Jekyll folder that I use to run my website.

Now, as the trust fades with each word, should I be concerning myself to make a Git-like backup to track a very detailed version history? Will I need to show others proof that something I wrote was slowly and meticulously crafted with dozens of mistakes and loopholes before sharing the final drafts? I would also need to have the right timestamps and metadata to validate my work. Will we see someone offer Blockchain as a solution to track the validity of building each story page by page?

While we wait and see how the world behaves, as a last thought, let me share a recent study caught my eye - more work by scientists from the Global South is getting published, as LLM helps them write the articles more clearly. Link to the post

p.s. I wrote this on my Mobile, so expect some update/edit over the week.