While unbelievable feats of ChatGPT and Generative AI tools have been all the rage across social media, it’s impossible to appreciate the immense scale and transformative power of these AI products unless you try them yourself. And very soon, you come to two undeniable realisations about your own limits as a human being. How small and limited you are and how vast and limitless you can be.
Granted I didn’t feel that way on my very first day of using ChatGPT-4, chatting with the seemingly know-it-all AI chatbot, or when I started dabbling with DALL-E 2 or creating “Pixar caricatures” of everyone I could think of (inspired by Mithun’s exploits with the medium), but it isn’t long before it dawns on you. When it comes to making magic, Generative AI tools trained on LLMs (large language models) like ChatGPT, DALL-E, Midjourney or Stable Diffusion are only as good as their prompts.
Before you start to wonder what ‘prompts’ are, they’re nothing but instructions or commands given to a computer. Not unlike queries you input into Google or your phone’s universal search, for example. Every time you type something in Google’s search field, you are essentially instructing or commanding or prompting it to search for more information related to your prompt. In a way, it’s reassuring to think we’ve all been ‘prompting’ for information from the very second we searched for anything online – whether through a web browser or an app. Every time you search for a song on Spotify or a document in Google Drive, a location on AirBNB or an item on Amazon, you prompt your way through all those apps and services. It’s only when you’re searching for something generic that you turn to Google or Wikipedia, or even Alexa or Siri. An important thing to realise here is that none of these online apps or services work for you if you don’t ‘prompt’ them in the first place.
It’s much the same with AI tools like ChatGPT or Midjourney. All these AI tools are reactive in nature, which means they don’t start weaving their magic unless you ask them to do something. It can be anything. And how you talk to them, how you interact with them, ultimately decides what you can do with them. Just like the old adage that a chain is only as strong as its weakest link, ChatGPT or Midjourney or any other AI tool’s ability to respond to your prompt is only as good as the quality of your prompt. Let me explain this with an example.
If you prompt ChatGPT to tell you the best tips to write a project report, it will do just that – produce a list of tips you’ll need to follow. But if you cleverly feed information about your project to ChatGPT as part of your prompt, you can ask ChatGPT to write your project report for you. Understand the difference between the two scenarios and their respective prompts. It’s the same with AI art generation tools like Midjourney, DALL-E 2 or Stable Diffusion – there’s a difference of images and their quality between “a young woman alone on the beach” versus “a 1980’s polaroid picture of a young South Asian woman alone on the beach during sunrise”. When it comes to ‘chatting’ with AI tools, the prompts matter. Being specific with your prompts as much as possible matters, guaranteeing the best possible results – at least closest to what you desired, anyway.
As far as my own limited experience with using ChatGPT, DALL-E 2, Midjourney and Stable Diffusion throughout the past month (in helping put together Digit’s “Written by AI” April 2023 issue), I know my own ability to extract more out of these tools improved significantly as I started to get more detailed and specific in my demands of them. It was no walk in the park, believe me, as the team transformed into a hive of “prompt engineers” trying to find the perfect prompts for various text articles and their accompanying images.
In time, I have no doubt that future versions of ChatGPT and AI art tools will become more savvy in reading our mind. Instead of us fumbling with our creative blocks in trying to write the perfect prompt for something, these AI tools will “prompt” back, trying to either help us clarify our own thoughts or to present us with choices of creative possibilities. Perhaps, they’ll even reincarnate as much smarter, all-pervasive chat assistants like Siri, Alexa or Google Assistant that reside in your smartphone. With access to your emails, calendar, documents and photos, they’ll start prompting you for instructions – on how to respond to your boss’ email, on what gift to send to your mom who turns 60 next week, on how to file your taxes within the deadline, and many more things. Scarily, I look forward to that day becoming a reality very soon.
This column originally appeared in Digit’s April 2023 issue
