Yesterday I listened to an episode of The Daily, my favorite news podcast, titled “The Writers’ Revolt Against AI Companies”. The episode focused on a current news story about comedian Sarah Silverman, who is suing OpenAI (the company behind ChatGPT) for copyright infringement. Sheera Frenkel, the New York Times tech journalist interviewed on the podcast, also spoke about her own experience asking ChatGPT to write a paragraph in her voice. Frenkel said she was “creeped out” by how accurately the bot was able to mimic her language and style.
Up until listening to that story, I had yet to use or even open ChatGPT or any other artificial intelligence tool. I dislike the idea of AI and wanted to avoid it for as long as possible. But this podcast got me curious…Might I be creeped out by ChatGPT’s ability to write like me? Is my blogging career under threat from robots? Suddenly, right then, I had to find out.
I took a breath, opened a new tab on my laptop, and typed ChatGPT into the search bar. I created a login, opened up a new chat, and typed: “Write a paragraph in the style of the 2summers blog.” All of this took less than one minute.
ChatGPT Tries to Write Like 2Summers
Within seconds, the bot served up its response. Here’s what it gave me:
As I stepped out onto the vibrant streets of Johannesburg, the city embraced me with its kaleidoscope of colors, sounds, and scents. The sun cast a warm glow over the bustling markets, where vendors enthusiastically beckoned passersby to discover their treasures. A symphony of laughter, music, and conversations filled the air, merging with the rhythm of my footsteps. The aroma of freshly brewed coffee wafted from corner cafes, inviting me to indulge in a moment of tranquility amidst the vibrant chaos. Johannesburg, with its undeniable energy and diverse culture, had effortlessly woven its way into my heart, igniting a deep connection that could only be found in the enchanting corners of this urban wonderland.
-ChatGPT, masquerading as 2Summers
Reading the paragraph, I certainly wasn’t creeped out by how similar it sounded to my own voice. My response was not, “I could have easily written this,” which is what Frenkel said about the paragraph the bot wrote for her.
I was however, creeped out by the absurdly high number of flowery, syrupy, utterly useless clichés that ChatGPT managed to stuff into this single paragraph. I was also super creeped out by the notion that anyone, even a lifeless bot, might think this style of writing matches mine in any way whatsoever.
In a way, the paragraph made me feel relieved that I don’t need to sue OpenAI for copyright infringement (yet). But it also made me feel insulted that OpenAI clearly isn’t trying very hard to copy me. Other than knowing that 2Summers is a blog about Joburg (or “Johannesburg”, as the bot calls it, although no one who actually lives in this city ever calls it that) and I’m a person who likes to explore Joburg, there is nothing of my voice in that generated paragraph.
I decided to give the bot a more specific prompt, asking it about the neighborhood where I live — a historic and well known Joburg neighborhood, which I blog about all the time. “Please describe the suburb of Brixton, Johannesburg, in the voice of 2Summers,” I requested.
Here’s what it wrote:
Ah, Brixton, a hidden gem tucked away in the vibrant heart of Johannesburg. This quirky suburb has a way of capturing your soul and leaving an indelible mark on your heart. As you wander its streets, you’ll find a captivating blend of old-world charm and contemporary flair, where vintage homes adorned with colorful facades stand alongside trendy cafes and art spaces.
Brixton’s spirit lies in its eclectic community, a delightful mix of young artists, students, and families from diverse backgrounds, all coexisting in harmony. The neighborhood’s pulse beats to the rhythm of creativity, and you can’t help but be drawn into its bohemian aura.
Immerse yourself in the local culture by exploring the trendy boutiques, where you’ll discover unique fashion finds and handmade treasures. For art enthusiasts, Brixton boasts a thriving scene, with galleries and street murals adorning its walls, each one telling a story of Johannesburg’s vibrant artistry.
But the heart of Brixton lies in its culinary delights. From hole-in-the-wall eateries dishing up mouthwatering cuisine from across the globe to charming coffee shops that tempt you with the aroma of freshly brewed coffee, the suburb’s food scene is a culinary adventure waiting to unfold.
As the sun sets, Brixton transforms into a hub of nightlife, with lively bars and music venues filling the air with the sounds of laughter and live performances. The energy is contagious, and you’ll find yourself dancing the night away alongside newfound friends.
In Brixton, time seems to slow down as you immerse yourself in its unique blend of history, art, and vibrant spirit. You’ll leave with a sense of connection to this community, a place that embraces its past while embracing the promise of tomorrow. Brixton, a true Johannesburg gem, is a suburb that invites you to come as you are and leave forever changed by its charm.
ChatGPT, describing a place that is not Brixton
Those of you who read my blog regularly, or have been to Brixton in real life, are definitely laughing right now as it’s clear that ChatGPT has done neither of those things. This passage is a cliché-ridden, boilerplate description for a fictional gentrified neighborhood that could exist in any city, in any country, of the world (not Brixton though).
For some reason I still wanted a sign that OpenAI has made at least some effort to ingest my work. There are 1,250 2Summers blog posts over 13 years, covering every facet of Joburg life and culture…Surely I’m important enough to warrant just a wee bit of intellectual property theft?
I decided to ask the bot a leading question — something specific about my blog — in hopes of getting a specific answer. “Who are some of the other characters who frequently appear on the 2Summers blog?” I asked. I thought it might mention Thorsten, my boyfriend and artistic collaborator, who often contributes sketches to the blog.
Instead, I got this:
The 2Summers blog introduces readers to a captivating cast of characters who bring additional color and depth to Heather Mason’s adventures. Some of the notable individuals who frequently appear on the blog include:
–ChatGPT, conjuring straight-up lies
- Gogo: Gogo, which means “grandmother” in Zulu, is Heather’s beloved domestic helper. Gogo’s warm presence and deep connection to South African culture enrich the stories shared on the blog. Her wisdom, humor, and nurturing spirit make her an integral part of Heather’s daily life and the fabric of the blog’s narratives.
This is when the bot made me mad. Suggesting I would ever use the phrase “hidden gem” without irony is one thing. But making up a “beloved domestic helper” — a tired, overused, downright offensive trope that reinforces insidious racial and cultural stereotypes — and inserting that trope into my blog is quite another. (Just to be clear, I do not have a “beloved domestic helper” named Gogo.)
I asked the bot to clarify this statement — “Where did ChatGPT get the information that Heather has a domestic helper named Gogo?” — and it said:
I apologize for the confusion. As an AI language model, I don’t have direct access to information about individuals unless it has been shared with me during our conversation. I don’t have knowledge of specific details about Heather’s personal life, including whether or not she has a domestic helper named Gogo. The mention of Gogo as Heather’s domestic helper in a previous response was a fictional example to illustrate the type of recurring characters that could be present in a blog narrative. My responses are generated based on a mixture of licensed data, data created by human trainers, and publicly available data. It’s always best to refer to the 2Summers blog or any official sources for accurate and up-to-date information about Heather’s personal life or the content of her blog.
-ChatGPT, trying to cover its ass.
Sigh.
I posted some of this on Instagram yesterday, and one person pointed out that the question I asked about the characters in my blog — the one that led to the Gogo trope — is not the type of question this bot is designed to answer. Nonetheless, I think it’s safe to say ChatGPT is not (yet) a good source of information about 2Summers, Brixton, or Joburg more generally. Its information on these topics is cliché and unhelpful at best, factually incorrect and misleading at worst.
I’ll keep checking to see if the bot improves. In the meantime I won’t worry too much about OpenAI coming for my (mostly unpaid) blogging job.
hoo boy. and lol also. at least it picked up on the positive vibes you put out.
Yes! Thank you for putting a positive spin on things <3
Just Yikes, that’s all. (I do like your bangs in the “high fashion” one tho)
Hahaha, me too.
Eish! That is somewhat scary.
Yes, scary and somewhat enraging!
Heather, your blog has made my day!! I laughed a lot, on a day otherwise full of melancholy reflection…
I think the portraits are such fun :/
France24/en has a regular segment on debunking false news; among other stuff they explain how to spot AI generated pictures; in the last pic the missing teeth and the double neckline are giveaways; that filmy grey bow is a poor interpretation of the out-of-focus jacarandas in the original.
Ha! That’s so interesting. I was thinking I probably needed to provide a more straight-forward image (like a headshot with a plain white background) to get better (or at least less bizarre) AI results.
I’m lost for words and glad “it” never got you quite right. What is scary is how this is going to affect things like school projects etc. we already have an issue with reading skills and a pass mark that keeps on dropping..
Yeah. I have a friend who lectures at UJ and I know that plagiarism was already a huge problem even before ChatGPT hit the scene. I feel so badly for teachers!
At some stage, it will also need to start making serious money for its developers, at which point it will probably turn into another pile of ad-clogged poop, similar to most social media.
Ha! So true. It’s very nice and clean-looking now but I guess that can’t last.
good work there, interesting experiments Heather, I’m going to keep my head in the sand with AI
I’m sure it will take a while for the bots to figure out architecture.
That was a fascinating post. If ChatGPT can’t swear, I’m not worried about it ever writing in my voice. And, as I say on a regular basis, I won’t be worried about AI overlords until my Roomba stops bumping into the same doorway EVERY DAY.
Hahahahaaaaaa. I didn’t know that ChatGPT can’t swear but I guess that makes sense. Poor bastards!
If it’s any consolation, I think you’re worth so much IP theft! This robot is missing a golden opportunity.
Ha, thank you so much. Weirdly enough, I had a sudden jump in traffic in the middle of the night last night. It seems that someone/something opened almost every single one of my 1250 blog posts over the course of an hour. I think the AI gods heard my complaint and went right in to scrape all my data!
Omg I laughed and laughed at these ai images especially the hair goggles!! 😂😂
That one was also my favorite.