Review: ChatGPT’s Satirical Ability

I asked ChatGPT to write its own Onion-style article with the same title as what I published yesterday.

Make your own judgement on how it did, and continue on if you’d like to read my thoughts.¹Disclaimer: I commit the cardinal sin of explaining a joke later on in this post.

A major critique I’ve seen of ChatGPT’s prose is its tendency to be formulaic and predictable, as if its only resource on essay-writing was a handout from a tenth grade English class. It just so happens that the average person, at best, writes in one of these five-paragraph type of formats and never graduates beyond it. This is the beauty of machine learning: when most of the population can only do things at a certain level, it’s hard for the machine to rise above that because the masses provide significant input on what’s acceptable.

I think ChatGPT’s version is indeed totally acceptable. It uses satire effectively, tossing in jokes that make it clear what we’re poking fun at, including a few general sentiments that mirror my own. For example, “It’s refreshing to see an AI that isn’t constantly trying to take over the world or enslave humanity” has a very similar spirit to one of Ron Anuba’s statements in my version. The format is also similar to begin, where we introduce the name of this new AI and gain quotes from both it and a human, establishing the world this satire is representing.

In this way, both ChatGPT and I were successful in recreating the AP-style articles that The Onion writes: brief news reports written in a neutral voice, allowing the facts and quotes therein to do the heavy lifting of the satire.

Yet the overall structure of the ChatGPT article is made uninteresting due to the lack of genuine stylistic choices. This article has a simple format:

1. Introduction and establishing our character.
2. Supporting statement for the character.
3. Detracting statement against the character.
4. Rebuttal to the detraction.
5. Concluding statement.

A simple format does not imply simple thoughts nor a lack of style. Unlike myself or ChatGPT, an excellent writer can weave a complex web of thoughts within a basic underlying structure; in fact, this is likely preferred. There’s a reason we can break down stories²Hello, Hero’s Journey. into identifiable parts. It’s the execution that tends to make a difference.³This is assuming the writer is sufficiently competent to understand a structure they should be using.

ChatGPT’s execution lacks the small details to make an excellent article. In terms of The Onion‘s style, it misses the personality and specifics common to their writing. Instead of naming names, it went with “one programmer” and “a rival AI creator.” It also doubles up on fake facial expressions, using both a virtual grin and wink.

There are mechanical choices that feel clunky:

Despite the criticism, Opti remains optimistic…

Yeah, I would imagine it does! It feels nearly tautological and on-the-nose. An immediate rewrite that comes to mind is:

Opti does not let the criticism affect its outlook.

The final and most glaring issue I have with ChatGPT’s version is the lack of a punchline. Nearly every article in The Onion⁴Going all the way back to A Modest Proposal you still see this tendency. ends with a final twist of the satirical knife, a concluding statement that either adds an additional layer of complexity on the established world, or is simply the best joke the writer had to give as a closing statement.

ChatGPT offers a moral instead.

I could be generous and say including a moral as the final line in an article about an optimistic AI is satirical and twists the expected format. Even if that were the case, it is poorly executed and undercuts the entire article. If this approach were to work, it would have to be in the voice of Opti (or another character), rather than the reporter if only because the AP-style writing of The Onion disallows pretty much anything except repetition of facts and quotes. Instead we could change the closing line to something with the same sentiment in words, but with an intention of continuing the satire:

At press time, Opti was reportedly coaching a support group for developers affected by improvements of GitHub Copilot, reminding them that in a world where machines are becoming more advanced, there’s still room for a little humanity.

We offer up the sentiment for consideration, but considering the point of the satire is to ridicule the idea of an optimistic AI being created with these goals in mind, we bat it down just as quickly.

I think ChatGPT is a neat tool, and whenever the next version(s) of GPT come out we’ll be in for new whirlwinds of fuss over it. Speaking about the ethics of such a tool is far beyond the scope of a bonus Tuesday post, but I’ll leave it here: the best use of these tools that were trained on improperly-obtained data⁵I say this with near certainty due to the complications of doing anything else. is inspiration. If you’re using AI art, have it create concepts that you can give to a real artist to execute. If you’re stuck on a topic, why not give ChatGPT a ring? It’s the same service I perform for a friend stuck on a sentence or a few ideas that they want to string together. Doing pretty much anything else is in poor taste, and only contributes more mediocre content to the internet.

• 1
  Disclaimer: I commit the cardinal sin of explaining a joke later on in this post.
• 2
  Hello, Hero’s Journey.
• 3
  This is assuming the writer is sufficiently competent to understand a structure they should be using.
• 4
  Going all the way back to A Modest Proposal you still see this tendency.
• 5
  I say this with near certainty due to the complications of doing anything else.