How to Use ChatGPT in Non-Evil Ways
Yes, AI may destroy the world as we know it. But it can also make your working day a hell of a lot easier.
Everywhere you look these days, there are horror stories about AI. It’s coming for our jobs; it’s telling lies; it might destroy the Earth and wipe out humankind. But does any of that really matter if AI can also make your work day a little easier?
In a world where work is constantly evolving, there's a disruptive force making its mark in offices everywhere: ChatGPT. In fact, a recent survey found that 43% of working professionals have used AI tools such as ChatGPT to accomplish tasks at work. Of these, nearly 70% admitted they hadn’t told their boss about their virtual assistant (which perhaps shouldn’t come as a huge surprise).
Some might say this is all slightly dubious from an ethical standpoint – getting someone else to do your work for you is a bit of a no-no, even if it’s a robot. But in our contemporary hustle and grind culture, ChatGPT’s popularity makes sense. When productivity is prized above all else, who wouldn’t want to streamline their to-do list? ChatGPT is basically the latest shiniest efficiency hack.
Need help scheduling meetings, booking flights, or ordering office supplies? ChatGPT's got your back. It can also generate code, brainstorm ideas, and write emails in a few seconds flat. But where to start? How can you harness the full potential of this futuristic assistant? This is where the experts come in. In the interest of your work emails and uni essays, VICE has chatted to ChatGPT gurus about how to get the most out of it. It's time to ride the ChatGPT wave into a brave new professional frontier.
Don’t believe me? Well, that whole intro was written by ChatGPT. Or was it? Who knows what’s real and what’s AI-generated any more – and if you’ve got a work deadline, who cares?
What is ChatGPT?
Curious how ChatGPT performs its digital wizardry? Let's demystify it for a moment. The program uses a language model trained on a colossal dataset, so it can predict and generate text based on the patterns it has learned from the training data. Ask it a question or tell it to perform a task, and it doesn't just parrot back your words; it generates answers that are eerily human-like.
OpenAI, an American AI company, launched ChatGPT in Nov. 2022. (They’re also responsible for Whisper, an automatic speech recognition system, and DALL-E 2, the AI art generator that briefly took over Twitter with its unrivalled meme generating powers.) There’s a paid version, ChatGPT Plus, but the original is still free, meaning anyone and everyone can use it. And everyone has been using it – for everything from writing scientific research papers to poetry to asking it what to make for dinner.
Raven (who is using a pseudonym for privacy reasons) is a software engineer who worked at an “unnamed big tech company”, where his team did experiments with ChatGPT. “We were essentially playing around with prompts,” he says. “We started with prompts like: ‘The following is a conversation between a person and Anthony Bourdain’,” he explains, “and then you'd speak to Anthony [via ChatGPT] about food, and it would emulate his tone of voice.”
Then Raven went one step further. “We built this thing that was a bit like a video game character builder,” he says. “So you'd be like: ‘The following is a conversation between an AI who's 50% Anthony Bourdain, 25% Nigella Lawson, and 25% Keith Floyd’ – and you’d say that it's an AI assistant for a 28-year-old who lives in London and likes cooking.” The idea, says Raven, was that you’d use it via your phone in the kitchen, and ask things like, ‘What should I make for lunch?’
“Obviously it's just sort of, ‘mega auto complete’,” he says, but when he tried it, strange things started to happen. “I asked it what I should have for lunch and it answered, ‘You could make tzatziki’. And in my fridge – as a 28-year-old male, living alone – it just so happened that all I had was some yoghurt and a cucumber… It was pure chance, but it was very eerie.”
Undeterred, he rigged ChatGPT up to an audio system, so he could talk to it more casually and it could reply. “If you did that over the course of a year, you could start data mining yourself,” he says. In theory, he says, you could train it to be nicer to you at certain times of the year, “so if you had seasonal affective disorder, you could tweak it in winter so it was a bit nicer to you.”
Raven continues: “Then I was like, damn, I'm single and I’m basically living with ChatGPT.” So, yeah: Experiment all you want, but maybe don’t go that far.
How do I use ChatGPT?
So we know it all starts with a prompt. Basically, you throw a question or request its way and it'll serve up an answer. It's like texting a wise friend. But wait, there's a catch – a good prompt is the key to unlocking the magic. Why? Because ChatGPT's brilliance lies in its ability to generate responses based on the context of your request. And also because it seems to have a habit of just making stuff up if you aren’t vigilant.
There’s a lot you can do to get the best possible results. Firstly, make your prompt specific and detailed. The more you tell ChatGPT, the better it can help. In a work context, this might mean spelling out both what you want it to generate (an email, an article, or the bullet points for a presentation) and the style or tone of voice you want it in (is this for your work husband, or your boss?)
For instance, if you need help brainstorming ideas for your project, a good starting prompt might be: ‘Help me brainstorm creative ideas for our upcoming marketing campaign for an artist who makes hand-painted plant pots,’ rather than vague prompts like, ‘How can I sell more plant pots?’ If your prompt is unclear, it's the equivalent of telling a human colleague, ‘Just do something!’
Basically, there’s an art to prompts – indeed, writing good ones has become its own highly-paid job role, known as prompt engineering – but for the newbies and amateurs, the best thing to do is experiment. Talk to it like a person, and feel free to ask follow up questions. Keep tweaking your prompts until you get what you’re after.
Raven gives one real, and very helpful example. “Having spent half a year doing ChatGPT experiments, I, of course, wrote my resignation email to my manager using it,” he says.
To get the tone right, he gave the bot quite a lot to work with. “The prompts were like: ‘Such and such is 28-years-old, and is leaving their job at [name] agency. It was a good company, but they've been offered something elsewhere, please write a nice resignation email. Keep it sweet, and hint at being available for any freelance contracts in the future’,” he says. Needless to say, Raven concludes, “It did a very good job.”
Okay, so what else can I use ChatGPT for?
Well, that’s the million dollar question. Raven’s already given one great example, but he has another up his sleeve. After leaving his job using ChatGPT’s resignation letter, he then got a new job by using it again. “I used it to co-write my cover letter for my new job,” he says, “and it was a gift from God when writing my CV too, where I had to change all the tenses.”
These are the kinds of fiddly, time-consuming jobs that ChatGPT excels at, according to Johnny, an AI academic and university lecturer, who is also speaking anonymously. Johnny uses ChatGPT to write his academic articles “explaining technical things,” he says. “I will often reword or rework these, sometimes very significantly,” he adds, “but I find it much easier having an example to work from and adapt than writing from scratch.”
He also uses ChatGPT to write code for his student exercises. “It’s really good at this,” says Johnny, but when I put that to Raven he disagrees. “We got told not to use ChatGPT to write any code,” he says, “because even though it can do simple things quite well, if there was a little thing that got slipped into production that went wrong, all hell would break loose.”
What about more run-of-the-mill office tasks, like writing a presentation? “You could give ChatGPT a bunch of bullet points and it would write a really good sentence,” Raven says, “and you could tweak it and be like: ‘I'm presenting this to designers, so make it interesting for designers’, or ‘I'm presenting this to someone higher up, so make it sound more formal and sensible’.”
Okay, but what about one of the most daunting tasks of all: You've got an uni essay due yesterday. ChatGPT to the rescue, right?
Unfortunately, Johnny puts his university lecturer hat on and slams the brakes on that one. “Anything written with ChatGPT is very obvious to spot when you’ve seen it a couple of times,” he says. “It has a very specific and predictable way of structuring written texts.” There are some other tell-tale signs a student has had “help”, he says. “Students who you know can’t speak a word of English will suddenly hand in essays with no spelling errors or grammatical mistakes… Or you’ll receive essays where there is a spelling mistake in the title or file name but nowhere else in the whole 2,000 word document.”
Quite often, it’s the references that are a giveaway. “ChatGPT-written essays will have references that are unrelated to the actual topics of the essay, and include citations that aren’t in the essay at all,” says Johnny. “The worst offender was a student who submitted an essay with references to papers that didn’t even exist.”
All that being said, Johnny still admits that he uses ChatGPT to even help with marking assignments, including “using ChatGPT to write feedback telling students not to use ChatGPT…” What a mixed up world we live in.
In essence, while ChatGPT might be the Swiss Army knife of workplace AI, it's not infallible. It's essential to use its suggestions as a starting point, not gospel. After all, even AI has off days. But if you stick to simple but time-consuming tasks like CVs, cover letters and tricky work emails, you might just be golden. “ChatGPT has done absolute wonders for my career,” says Raven, “for real.”