Using a Language Model to Write My Blog Would Be Disrespectful to My Audience and Worse Than Pointless

I have been encountering an increasing amount of generated writing online and in other personal and professional communications, and I want my audience to be assured that I will never subject you to the same.

I find it infuriating to realize halfway through a blog post that I'm reading LLM output. I'm not interested in reading LLM output, refined or otherwise, and I know others feel similarly.

I am not going to go so far as to say that there is no place at all for LLM-generated prose¹, but I will claim that all generated prose should be labeled as generated. Authors who do not indicate when prose is LLM-generated are not honestly representing themselves or their work.

As this applies to my blog:

If you're here, I greatly appreciate the time that you're spending reading my writing. Your time is not worth less than mine. If I could not be troubled to pick and choose each and every one of my words, why should you bother to read them?

Of what use would this blog be to me if I were only to use it to recycle ideas through a language model? My thoughts would only be muddied by the random choices of the next-token predictor.

One of the great benefits of keeping a blog, or journal, is the way writing about topics of interest help you crystallize your opinions and work out contradictions in them, and to refine ideas. Many writers have remarked on this. This blog is hardly a personal diary but generating entries for it would feel much the same to me as generating a diary entry. How am I supposed to learn from reflecting on my thoughts if the LLM does the reflection?

Then there is the issue of style. Using an LLM to "aid" in my writing, would be to unceremoniously smear the styles of countless writers over the top of my own, resulting in an off-putting chimera² inevitably worse than either my own or those which have been copied, or ripped off, depending upon your perspective.

Aside from being tremendously boring, representing generated ideas as my own in a context like this blog would be borderline fraudulent. I struggle to see the practical difference from plagiarism.

Legally, it seems the powers that be have decided there is absolutely nothing noteworthy about prompting an LLM for some text and then publishing it as though it was written by the person who generated it.³

Socially, we should reject that. Presenting LLM-generated text as human created is a lie. "I wrote this" means "these are my thoughts" and if you generated it, they are not your thoughts and it is a lie to imply the writing is your thoughts instead of a next-token prediction, even if it is a very good, thorough, and useful next-token prediction!

It should be seen socially as shameful to replace your own thought with the output of an LLM, especially when using that output to bamboozle an audience into believing the output was genuine and novel thought from a fellow meatbag. It cannot be proven if text is or is not generated, but if you catch your friend, your spouse, your mom, using ChatGPT to write a text message, a note in a birthday card, or a blog post, scold them! That is an amoral act at least on par with not returning your grocery cart to its corral.

Passing off generated fiction, personal correspondence, advice, or even opinions as genuine writing commits the same lie which LLM psychosis sufferers have fallen victim to:

the reader is misled into believing that they are reading genuine insights from the experience of another living breathing conscious being, when in fact they are reading a randomized approximation of a million Reddit threads.

Would you do that to someone you care about? Replace your carefully chosen words to them with a poor approximation of your feelings laundered through a filter of Redditors?

My most valuable audiences will speak to me in person. They will know if my ideas are mine, and will discover if my eloquence is really the clanker's.

It would not be a sustainable ruse even if I thought LLMs could improve the quality of my writing which they can't because…

I like writing. I'm good at it, or at least decent. If I didn't like writing, why would I have a blog?

Software is different. That's not to say these issues aren't important, and I will continue to try to call out the bits of code I share on this blog as being variously generated, artisanal, or something in between.

But I think the discussion surrounding LLM use for software development is fundamentally different than the one around using them to write text for human consumption. I have not fully developed my thoughts on that topic, and will leave it there for now.