3 ChatGPT Truths the AI Bros Don't Want You to Know

If you spend time on the growing cesspool that is Marketing Twitter, you’ll find two things:

1. People claiming they’ve replaced their entire marketing team with ChatGPT and that you should do the same.

People selling ChatGPT courses and prompt templates with the zeal of an amphetamine-amped crypto bro from 2021.

If you’re a marketer or content creator, it can feel like you just stirred out of your holiday food coma and—BOOM—AI has taken over. It’s a new world, and you have two choices: Accept a life of destitution, or do the jobs of 57 people with the help of your new generative AI friend.? YOU HAVE UNTIL MIDNIGHT TONIGHT TO CHOOSE.?

But is this reality or just an illusion created by the engagement industrial complex?

On one hand (picture this hand gesticulating wildly with a lot of New Jersey energy): GPT-3 tools—like ChatGPT, Jasper, MidJourney—are incredibly dope. Mind-blowing new apps and use cases are popping up every day. I’m jazzed both as a content creator and as a marketing leader at A.Team. (For instance, MidJourney made the hero image of the sad cyborg at the top of this newsletter.)

On the other hand (picture this hand with serious eyebrows drawn on it in Sharpie): Misusing ChatGPT could get you fired because spoiler alert: ChatGPT lies. A lot. The layer of hype over ChatGPT is so thick right now that it’s easy for most people to misunderstand it.?

In this week’s newsletter, I’m going to embrace my new-dad energy and give you all the big warnings about the limitations of ChatGPT and GPT-3. Then next week, I’ll give you the hype-free breakdown of the use cases that excite me the most.

ChatGPT lies a lot and doesn’t actually know what it’s talking about?

The first time I used ChatGPT, I asked it to give me 10 tips on how to navigate procurement inside enterprise organizations. It did pretty well—imitating the work of a C+ freelance writer with solid subject-matter expertise who had forgotten to cite sources.?

But in reality, ChatGPT—and other GPT-3 tools—have no actual idea what they’re talking about. Large Language Models (LLMs) like GPT-3 are text predictors—basically, the autocomplete you’ve grown accustomed to in Google docs but on steroids. They’re not search engines. They’re not databases. They’re just really good at guessing what comes next—so much so that the answers seem like magic. But it will also create convincing falsehoods. As AI expert Gary Marcus explains:

Fundamentally, what they learn are relationships between bits of text, like words, phrases, even whole sentences. And they use those relationships to predict other bits of text. And then they do something almost magical: They paraphrase those bits of texts, almost like a thesaurus but much much better. But as they do so, as they glom stuff together, something often gets lost in translation: which bits of text do and do not truly belong together.

ChatGPT is dangerous to use as a 1-1 replacement for Google. It will totally fabricate someone’s biography.

It will get basic facts wrong, as much as my liberal heart wishes they were true.?

And it is SUPER confused by what is and is not a mammal.

You need to proceed with a ton of caution when relying on ChatGPT for research and content production. Tools like YouChat are trying to solve this problem by combining a GPT-3 chatbot with search, but in my experience, it still returns a lot of confusing or off-base citations.?

Replacing writers with AI is dangerous and could get you fired

I see a lot of AI hustlers proposing that smart marketers will just delegate writing content to AI.?

CNET—the once-renown tech outlet—tried this. The result: widespread factual errors and rampant plagiarism, as Futurism reported this week. It’s the type of plagiarism you’d expect from a lazy 6th grader.?

Blindly publishing AI-generated content is just really bad for the world. The internet is already riddled with enough misinformation. If every website gets flooded with it, we’ll need Google’s algorithm to get a hell of a lot better at identifying the best sources of information.

At least 500 content marketers and journalists are going to get fired for AI misuse in 2023. Don’t be one of them.?

AI can’t write as well as people claim?

AI can’t introduce new ideas. It can’t conduct analysis or make direct comparisons. It can tell stories with the right prompt, but they’re pretty boring. (Yes, even if you “prompt engineer” it to tell a story in the style of Matthew McConaughey.)?

ChatGPT’s writing capabilities are impressive, but mostly because you’re going, “Wow, an AI wrote this!” When ChatGPT’s words are placed under a human byline, they curiously fall flat, as this Atlantic article cleverly shows.?

I’ve also seen a ton of hype over how tools like Jasper.ai and copy.ai can write ad taglines for you. I’ve tried them off and on for months, and while they’re useful for brainstorming new angles, the copy itself is pretty bleh. Bottom line: If you want truly good copy and stories, you still need that human touch.

With all those caveats out of the way, we’ll spend next week looking at how GPT-3 can turn talented content creators—like you, hopefully—into super creators. (For instance, MidJourney made the dope af hero image above!)

Subscribe below to make sure you get it.?