ChatGPT: friend, foe, or indifferent tool?

D’know early adopters? I’m not one of them.

In fact, I probably fall in the grey area between mass and late adopters, waiting for the fuss and hysteria to die down before dipping my toes into whatever may be ‘on trend’. And, because I like to stay ‘on brand’, I’m probably a little late to deliver my 2 cents on AI tool, ChatGPT. But, with the exception of this LinkedIn post I wrote during the height of the frenzy (remember that time you couldn’t get onto the platform?), I’ve taken the last few months to form a proper — albeit flexible — opinion.

Below I cover everything from the basics of what ChatGPT is, prompt engineering, what I’ve found it works best for in my portfolio, and whether this is the ‘end of writing’.

The basics (in layman’s terms)

Photo by ThisIsEngineering on Pexels

ChatGPT is an artificial intelligence (AI) tool that’s built on top of something called a large language model (LLM). LLMs are computer programs that are trained on giant sets of text data and, when asked a question, use this data to create a relevant answer in the correct context. As the name suggests, they are very, very large and, as a general rule of thumb, the bigger they are, the more accurate* they are.

*Accuracy issues are outlined in the section below titled “The bad”.

The GPT part comes from the type of AI architecture it uses, namely a generative pretrained transformer (GPT). Generative, because it generates a response to a query, Pretrained, because it’s taken developers tons of time and data to get it to do what they want, and Transformer, because this is the name given to AI models that can process sequential data (like words in a sentence, sentences in a paragraph, etc).

‘Chat’ is part of the name because it functions in the same way as a chatbot tool. If you ask it a question, it replies with a human-like response. This is possible because ChatGPT is also a type of AI known as natural language processing (NLP), that understands, summarizes, and generates content in human language (as opposed to code).

The good

Photo by srini on Pexels

As a writing tool/aid, it’s simply incredible. With the stroke of a few keys, you can generate ideas for content, find simple explanations for complex topics, and generate summaries and social media post content plucked from large bodies of text. It’s also the handiest of tools in transforming transcripts from interviews into coherent narratives that form the basis of long-form content*.

*See my guide on how to do this in the section titled “How I use it”.

Owning a calculator doesn’t make you a mathematician, just like using ChatGPT or Grammarly doesn’t make you a writer.

ChatGPT is not the first AI designed to generate content. (I know this because I’ve been fixing AI-generated blogs for ages already.) It is, however, one of the better tools we currently have access to. But, despite its obvious advantages, there are still some very real drawbacks.

The bad

Photo by Jan Prokes on Pexels

ChatGPT generates generic content.

It pulls data from its bank of knowledge and, even with the most carefully crafted prompts, it can pump out the biggest load of trash. It’s not capable of original thought so if you’re simply copying/pasting its half-baked outputs and using them verbatim beware.

I’ve found it can also get stuck in what I like to call the ‘loop of death’: every social post and blog can end up with the exact same format. I’d imagine it’s been programmed using templates for these functions, and even the combination of detailed prompts and desperate clicks on ‘regenerate response’ pump out nothing more than variations on a theme. Savvy readers can pick up AI-generated content and their opinion of you will drop.

Lastly, tools like ChatGPT can hallucinate, i.e., just make stuff up. Like anyone who talks convincingly about a topic they know nothing about, just because it sounds like it’s true, doesn’t mean it is. Plagiarism and fact checks are essential, lest you don’t mind your brand being damaged.

Granted this was not the most inspired prompt, but I got ChatGPT to do everything described above: Templated response, generic content, and hallucinate (I didn’t tell it the name of the company was “Plumb Fun”). Oh, is it just me, or is this not an engagement post.

The ugly

ChatGPT can make you LAZY.

Yoh — it’s SO tempting to just shove in a prompt and click repeatedly on ‘regenerate response’. I’ve caught myself doing this on several occasions (and you will too).

Granted, not everything that emerges from my brain is pure gold, but now there is some serious temptation to outsource actual thinking. Plugging in a prompt is easy, as is convincing yourself that the response is “good enough”.

Don’t forget that ChatGPT is not capable of original thought. Don’t succumb. Use your brain.

Engineering prompts

Photo by sol on Unsplash

I’ve been saying for years (about pretty much everything):

The quality of the output is only ever going to be as good as the quality of the input.

Prompts in ChatGPT are no different and, luckily, there are loads of resources available on how to craft the perfect input. A quick Google search will throw up templates and formats that you can use to ensure you’re inputting everything from the type of content you want, the tone and style it needs to be written in, the objective of the piece, and the type of action you want the reader to take after reading it.

Depending on what you’re trying to get ChatGPT to do, trying to craft the perfect prompt can be time-consuming and frustrating, leading to a giant waste of time and effort. Sometimes it’s just quicker to write the actual content yourself.

At the end of the day you’re telling a machine what you want and there is a lot that can get lost in translation. Case in point for “Plumb Fun”:

Using the prompt “rewrite this post in a more urgent tone” would have yielded a slightly better result.

How I use it

Trying to use ChatGPT to generate uniquely creative, out-of-left-field content (i.e., the type that gets noticed) is like trying to use a microwave to whip-up a Michelin-star meal. You’ll certainly get something, but it won’t tick the right boxes. That doesn’t mean that you can’t use it within your process, though.

My current niche is creating long-form thought leadership pieces for highly technical subjects (currently, I’m doing so for a leading company in the AI sector). The process of writing these articles involves interviewing subject matter experts from the organisation and transforming the conversation into a flowing piece of content, detailing specific aspects of the solutions they’re working on and the AI that’s being used in the process. Interviews are done conversationally and last around an hour. The transcribing tool (also a handy AI) spits out a document between 30 and 50 pages long, which is then used as the base for a piece that averages 1,500 words. The process of transforming the transcript into a coherent thought-leadership piece has been reduced by a whole day.

Here’s how I use ChatGPT to create long-form pieces that aren’t run-of-the-mill:

Step 1

Preprocess the transcript: Go through the document with an analytical eye, grouping similar concepts or themes together in a separate document. I usually start with headings that form a logical order or story and populate these sections with pieces of the transcribed discussion that fit the specific narrative.

Step 2

Do a basic edit of your initial draft: Step 1 usually results in a draft in the region of 5,000 to 6,000 words. I go through this carefully and remove any potential references to IP or information that needs to remain confidential. (Even though these articles are published in the public domain, I don’t want IP to become part of a data training set that’s used to refine ChatGPT.)

Step 3

Use ChatGPT to process your initial draft: You can’t copy your 6,000 words into the prompt box. Well, you can — ChatGPT can handle text blocks up to this capacity — but your output will be complete crap. Instead, take a few related paragraphs at a time and use ChatGPT to “transform this piece of transcript into a coherent narrative”. Copy/paste each output into a new document.

Step 4

Edit like your life depends on it: After step 3, your new document will still just be a steaming pile of dog poo, but it’ll at least have a basic structure that’s easier to work with. You can more easily identify repeated information and start moving text around to create some flow through the piece. (It’s here, too, that I usually add back any organisation-specific information or authorised IP, making sure it fits the context and flow of the piece.) Delete bloat to hit your word target, restructure sentences, correct grammar, and write section links as you would for a normal article.

Voila. You’re welcome.

Final thoughts

When ChatGPT was launched in November 2022, the big question was (and still is), “is this the end of human writers”? My opinion is, “not yet”.

As you can see from my transcript-to-article process, using ChatGPT is not a simple plug-and-play exercise. If you’re after decent content, it’s labour-intensive and still requires the human brain driving things from behind the keyboard. Will it remain that way? Only time will tell, but let me leave you with my final thought:

Since before the industrial revolution, advances in technology have resulted in large scale change for humanity.

From the wheel and printing press, to the steam engine, internal combustion engine, electricity, the telephone, and the internet, almost every new invention has been met with a mixture of fear and wonder.

ChatGPT — and other AI — is no different. It’s set to change the way we work, and those who embrace figuring out how to work with it will find themselves pulling ahead of those who resist.

Enjoyed this blog? Share it, subscribe to my monthly mailer, or connect with me on LinkedIn.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: