How ChatGPT Slashed My Medium Earnings To One Fourth
The true story of what happens when we stop trusting ourselves.
I have been writing on Medium since 2018. If you read my first article, you will find that I am not an English native speaker. Probably you have noticed it also reading this article. With time, I have improved, and yet, I do not trust entirely in my capacities, so I am still looking for new tools to overcome that limitation. And that is what took me to try ChatGPT and failing completely.
Writing is not only about orthography or grammar, there are things more important, like content, expressiveness and tone. And it turns out that some of them, even with the most well-crafted prompts, are outside the language model capacities.
Since I began to write, I tried several techniques to improve my craft. The first tip given to me came from Miquel Barcelò¹. He was a professor at my university, but also a book publisher. His tip was never publishing the same day that you wrote it. Instead, after writing something, we should keep it in a drawer for several days (or weeks), and then reread it. That was, without doubt, the best tip that he could give to anyone.
How it works is brilliant: when you write something, your writing and your brain are in sync. Every word has a meaning, not by itself, but because you know what to expect from it. But the reader does not share your state of mind, and that could leave some words with a little or no sense at all. If you wait enough, you will forget your words, and your intentions. At that moment, you will become another reader, and you will see what you missed.
Later came Grammarly. And it was the first hint of what would happen later with ChatGPT.
Grammarly is a tool to correct orthography and grammar. But it also goes one step further: it is able to suggest changes in style. And just because I am low confident in my capacity of writing correct English, I trusted more the Grammarly criteria than my own criteria. That was a mistake.
I discovered how wrong I was of trusting blindly in Grammarly corrections after watching the following video (probably it is the best video that I have seen about writing):
The video explains how to write in a very practical way, and avoids the more traditional school based on rules. The one that in fact follows Grammarly. If we compare both, they make two different styles, and we should decide which one to follow. I thought it would be hard to decide, but, later, in another video of the same professor, there was an active vs. passive sentence example that made me understand how huge my mistake was.
The example showed two sentences: “the cat is chased by the dog” vs “the dog chases the cat”, and the professor asks: “¿Which is the best?” The answer: “depends on the reader”. Grammarly will tell you that the second one is shorter, so it is more concise. Wrong! If you are a cat lover, you want to read what happened to the cat, so, if you read the active sentence “the dog chases the cat”, you have to mentally reorder it, so the cat becomes the subject. That simple example made me stop using Grammarly.
The next best improvement on my writing was by listening. Medium added the option to listen to your own articles, and one day I tried to listen how it would be read. Then, I discovered that I could improve even better the first tip. If the article sounded like it was read by a native English speaker, then it was ok. If not, I fixed until it sounded great. It was a sublime discovery.
And now we arrive at ChatGPT. Since ChatGPT has been trained with real English, from native English speakers, it is safe to assume that its English is more formally correct than mine. So I decided to ask ChatGPT to help me correct my English; instead, that is what happened:
Me: “I am writing a Medium article, and I need some help with English. Can you proofread the previous paragraph?”
ChatGPT: “Now we arrive at ChatGPT. As ChatGPT has been trained with real English spoken by native speakers, it is safe to assume that its English is more formally correct than mine. Therefore, I decided to ask ChatGPT to help me correct my English, and that is exactly what happened.”
Have you seen that? It has done a lot of changes. For example, it has removed the initial “And”. For me, it is important, and it turns out that it helps to focus better the reader. But for ChatGPT it is meaningless, and it removes it. The other major change is “real English spoken by native speakers” vs mine “real English, from native English speakers”. The first looks more correct, but the second has a special emphasis that is not present in the first one. In the second one, I can imagine you, the reader, feeling that being a native English speaker is something special, ¡it has its own sentence fragment! But the proofread version of ChatGPT does not have that emphasis.
So, because I was not happy, and I did not know how much of the Grammarly problem has been repeated, I tried other prompts. I even had long discussions with ChatGPT asking for corrections, reversions, whys, and more. And nothing helped. So I tried to narrow from proofread to just spellcheck:
Me: “I am writing a Medium article, and I need to spellcheck it. Can you correct grammar and orthography errors?”
ChatGPT: “And now we arrive at ChatGPT. Since ChatGPT has been trained using real English from native English speakers, it’s safe to assume that its English is more formally correct than mine. So, I decided to ask ChatGPT for help in correcting my English. And that’s what happened.”
Well, it does look better, right? Wrong! Ok, the first “And” is kept, but it made random changes here and there. In fact, the original response also says: “I corrected a few grammar and orthography errors, and also made some minor tweaks for clarity and readability.”. And I tried, and tried again to avoid those “tweaks”, and nothing was resulting ok.
My final impression about ChatGPT is that it has its own model about how English should be written. We can say that it has its own style, and it infuses its own style in everything that it writes. It does not perform the task that you ask, it only writes what it decides that fits more correct in the situation. That is the combination of what you asked, and what its style is.
And probably, although it is technically correct, it does not feel right. Like old voice synthesizers sounded too metallic, too robotic, it looks like that ChatGPT feels too robotic to our brain.
I reread some of the articles that I proofread by using ChatGPT, and I wanted to rewrite them again. In fact, even sometimes I had problems trying to understand what was written in them. It turns out that when I was reading the ChatGPT responses, I was using my own writing as a context. My brain was completing the ChatGPT context and I did not see the lack of proper meaning. I was failing in the first tip!
So, just because I did not trust my criteria about writing, I was letting ChatGPT dictate what to write. That was a great mistake. Because the important is not focus on writing rules, but on readers!
And just as a divertimento, I leave you a summary of the article written in verse by ChatGPT.
Trusting tools to write with more flair,
but at what cost did it come to bear?
Grammarly was the first to be tried,
blindly trusted, but it misapplied.The tip from Barcelo proved to be wise,
Leaving words in a drawer before we revise.
With the option to listen to our own voice,
we could write better, a discovery to rejoice.But then came ChatGPT to the scene,
to proofread with language real and pristine.
Yet, its changes were more than we thought,
and our writing style came to naught.The tool proved to be less of a help,
leaving our work in a misfit yelp.
So spellcheck was what we asked to do,
to keep our writing style all anew.
Thanks for the read. I usually like to write stories to think about how we understand and apply software engineering, and to make us think about what we could improve. If you liked the article, check my most successful stories on Medium to read more, or my list of stories.
I leave this other article here. It explains how a lot of “authors” have started to write articles on Medium using ChatGPT. Apparently correct, but completely lack of life. Given that the article author has a publication, he describes how his job has got a lot of more complicated, and the overall quality has lowered.
Is there any other article worth citing? Any experiences to share about?
- Miquel Barcelò was very well known in Catalonia in the circle of Science Fiction and Fantasy. He had contacts, and even friendship, with many of the most prolific Science Fiction writers at the time, like Vernon Vinge, Brandon Sanderson, or Orson Scott Card.
