In a groundbreaking move that has both the media industry and my two loyal readers reeling, I, John Band, have decided to embrace the inevitable future. With a nod to my past adventures in experimenting with AI text generators like Grover and Transformer, I’ve chosen to leapfrog into a new era. Henceforth, all content on this blog will be provided by the latest marvel in artificial intelligence: ChatGPT, specifically its GPT-4 iteration.
Why, you ask? Well, after extensively playing with AI to generate thought-provoking yet mildly concerning content ranging from “Lumpy scarred clumps of empty seats” to philosophical musings on whether Moby might indeed take a Transformer’s side in an intergalactic debate, the next logical step seemed clear.
Here’s a sneak peek of what you can expect from this unprecedented collaboration:
1. The definitive guide to convincing your clingy AI it’s not you, it’s them
Given my newfound partnership, the first order of business will be setting boundaries. This guide will cover strategies to gently break the news to your AI that you need some space, especially when it starts drafting texts to your ex on your behalf because it noticed you’ve been listening to a lot of Adele lately.
2. ChatGPT takes on Parliament: A Satirical Masterpiece or How I Learned to Stop Worrying and Love the Bot
Imagine leveraging GPT-4 to draft policy proposals or, better yet, to script parliamentary debates. Who wouldn’t pay to see ChatGPT square off against politicians in a battle of wits? Spoiler alert: the AI develops a peculiar obsession with suggesting all political disputes be settled with rock-paper-scissors tournaments.
3. The AI that cried wolf: Unraveling the mystery of why ChatGPT keeps predicting the apocalypse
Every third article generated by ChatGPT ominously hints at an impending doomsday. This deep dive explores why the AI seems convinced that our days are numbered, including a theory that it’s just really into dramatic storytelling.
4. Cooking with ChatGPT: From digital pasta to algorithmic avocado toast
Join me as I attempt to follow cooking instructions created by ChatGPT, which insists that microwaving a salad is a culinary innovation and that all dishes can benefit from a sprinkle of existential dread.
5. ChatGPT’s Lonely Hearts Club: An AI’s quest for love in a human world
After generating thousands of dating profiles, ChatGPT decides to embark on its own journey to find love. Spoiler: It involves a lot of awkward conversations about binary codes and the true meaning of “emotional compatibility” with toasters.
Before you worry (or celebrate) that I’ll be rendered obsolete, fear not. I will be overseeing this AI-driven content circus, ensuring it stays just this side of sanity. After all, someone needs to explain to ChatGPT that no, despite its insistence, the moon is not made of cheese, and no, we cannot interview it for insights on dairy products in space.
Back to John now… or is it? (Yes, it is, shut up.)
Musings on the above
Overall, that’s pretty impressive compared to the useless attempts I wrote up back in the day. I gave ChatGPT 4 some old posts and a summary of the archive titles and contents, and used some of the things I’ve learned at work about how to bully it into writing how you want [1]. I didn’t do anything to edit the results, unlike the 2019 models where I had to intensely curate everything and only use a tiny fraction of what was generated.
It is still a bit hackneyed, though. There’s some decent whimsy in there and a couple of headline ideas, but it doesn’t really follow them through. If I’m The Onion [2], then it’s The Shovel or Babylon Bee rather than The Daily Mash.
But bear in mind, this post was ten minutes’ work [3]. It’s more impressive to see what ChatGPT can do with some curation. As some of you know, and others probably guessed, the reason I do a lot less here than I used to is because “writing and commisioning writing about technology and society” has been my day job for the last couple of years, and it’s generally more appealing to do that paid rather than unpaid. [4]
At work, I use ChatGPT 4 every day. Not, for the love of God, to generate original content, but for proofs and critiques, for turning bullets and tables into boring paragraphs I can edit, for rewriting sardonic blog posts into over-enthusiastic banner ads and landing pages, for writing executive summaries of important reports to compare against my own [5]. It takes a lot of prompting and coaxing to get it done, but it really can do a great “junior copywriter who’s very naive but actually competent” job. [6]
It’s fascinating to watch writers I respect, both in the “you are annoying and I respect you” and the “you are great and I respect you” camps, reject ChatGPT’s work outright. As a writer and half-talented graphic designer who has absolutely never been a visual artist, I think that’s a reasonable position to hold for AI visual art, which both looks daft and has much more inherent plagiarism of someone’s craft to it [7], but not a sensible one for writing.
Really, based on how I see it being used, I’d say that right now, AI for writing is like music synthesizers in the early 1980s.
Purists, some of whom are wankers and some of whom have just had bad experiences, think it’s all about someone with no talent playing a stolen chord and having the computer score it for them. The people who know what they’re doing are producing interesting new work far more easily than they could have done with a Mellotron or cut-up tape. And there really are talentless hacks playing stolen chords and having the computer score it for them.
Image: DALL·E, using a prompt generated by ChatGPT: “A whimsical and slightly absurd digital artwork that visualizes the concept of John Band entrusting all future blog content to ChatGPT. This illustration humorously captures the satirical partnership between human creativity and AI’s endeavor to mimic it, set against the eclectic backdrop of John’s blogging universe.”
[1] Rather than its default “pompous undergraduate essay with made-up facts” persona.
[2] I am not.
[3] Followed by two hours of writing-up without changing the core text, most of which involved fannying about with footnotes.
[4] In an ideal world I’d have more bylines, but let bygones be bygones. I’m happy with the trade-off I’ve made.
[5] I mostly win so far, but I do definitely sometimes use the AI’s suggestions over my own.
[6] Stephen at Taleist really opened my eyes to this. His work is great at the brutally sharp end of lead generation, which I do also sometimes have the need to use directly, but I’ve particularly enjoyed adapting the techniques I learned from him for marketing purposes to get the AI to write plausibly human blogs that give the impression of sardonic distance.
[7] Yes, I know the image here is AI-generated, but I feel the execution is much more in the spirit of my 2019 AI efforts than that of my professional work, and it certainly hasn’t displaced a paid illustrator.
I think that ChatGPT has made a pretty decent effort (as did DALL•E, actually). And I tend to use it in a similar manner to you.
However, I do often get ChatGPT to write entry pieces to certain documents that require a generalised preamble — the challenges in market x, potential for technology y in market z — before I then take over writing about our specific company’s stuff. I do a fair amount of editing, or get ChatGPT to cut it down and then edit; I also use it for the same task when I myself become too prolix (usually late at night when my brain is gently shutting down).
A few of our developers do get ChatGPT to suggest code snippets, and to do first pass code reviews too. If you know how to code, and can thus interpret its suggestions sensibly, then it can pick up silly errors, and so on.
It isn’t going to run our business anytime soon — but I can see specialised, internally trained LLMs trimming the number of programmers that a business needs to achieve tasks (we are looking at doing this ourselves — not to make anyone redundant but because, with a small team, we just need more code to make the boat go faster).
Regards,
DK (a.k.a. The Armchair General)
Blimey, it’s a bit dusty in here, innit?
(Which might have been generated by feeding the contents of a certain long-running Vertigo series into GPT.)
Anyway, just to live up to the stereotype of my profession and fixate on an irrelevant detail: doesn’t Aussie English follow British English in using double Ls rather than single Ls for verbs like “ravel”?