This guy turned prompt engineering into a business
Good prompts are worth cash, Ben thinks
Hey Ben, I discovered your work online and found this interesting! I am actually a seller on Promptbase and I am super curious about it. I think that buying and selling prompts is an interesting idea, I have no idea if it will be a “thing” later on but I think it’s interesting to dig into it. Can you tell us who you are and how you ended up doing AI-generated art?
Ben: My name is Ben Stokes, I’m a programmer building tiny internet projects and writing about it on my blog Tiny Projects. I sort of fell into the AI generated art world by accident because of my most recent project PromptBase, which is a marketplace for buying and selling prompts for AI models.
PromptBase originally started off as a marketplace for GPT-3 prompts, but transitioned into a general prompt marketplace with the release of text-to-image models. So, I’m more of a GPT-3 prompt engineer than an AI artist! But, over the past few months, I’ve tested thousands of user-submitted text-to-image prompts for PromptBase and seen some incredible artwork.
Checking your personal website it seems you’ve been working on several small projects before doing this one (And I love it!). What are the things you learned that helped you building promptbase that fast?
I started my blog Tiny Projects a couple of years ago because I had a long list of ideas in my phone notes and I was frustrated with myself for not actually doing anything with them. My initial goal was to try launching one idea a week and just see what happens, but I quickly realised that was a crazy goal, and now try to at least build something once every few months.
PromptBase is the 8th project I’ve built on this mission. Generally I try to build things as tiny as possible to get it infront of users fast to validate the idea. If it works, I keep going with the idea, if not, I axe the project or sell it on.
With PromptBase in particular, I’ve built a couple of marketplaces before which made executing it a bit faster (one was strangely a marketplace for dentists!).
And how did you come up with this idea anyway?
My last project before PromptBase was called Paper Website, which lets you create a website or blog by handwriting out your page and taking a picture of it. It’s a strange idea, but I love this offline, distraction-free way of creating. I’ve nearly written 200 daily blog posts using this method.
A problem I had with Paper Website was that the tech which turns pictures of words into digital characters (OCR) does not work great on handwriting, it often makes a lot of spelling mistakes and jumbles up words. I’d heard of GPT-3, so I decided to see if I could use it to fix spelling mistakes.
Long story short, it worked! But, it was a very painful process for me to try and create a prompt to do this. I had to learn “prompt engineering”, and the first prompts I made weren’t great, which meant my OpenAI bill was pretty expensive. The whole time I wanted to buy a battle-tested prompt that I knew worked, but no GPT-3 prompt stores existed at the time.
After I launched my blog post about Paper Website, I had a random inquiry from an EdTech firm wanting to use my prompt to correct mistakes on test papers, which made me realise that well-written, reliable prompts are very useful and have quite a bit of value (look at Jasper, CopyAI etc. – they are built on prompts). So this is when I had the initial idea of a prompt marketplace.
It’s been a couple of months since you started and you even got featured on wired and TechCrunch. Could you share some data/insights about it?
PromptBase recently passed 10,000 registered users since launching in June, which is very exciting. The marketplace is growing each day with a healthy number of buyers and sellers. Some of our top creators have made several thousand dollars selling their prompts!
It’s been great getting press from such respected publications too. The idea of buying and selling prompts is a bit of a taboo topic, which has definitely helped PromptBase get even more press.
What are the people mostly trying to buy?
PromptBase prompts tend to be created for real-world use-cases, e.g. for creating icons, logos and stock photography – so the most popular prompts are centered around those themes.
There are some interesting trends though, for example, prompts that generate colouring book pages are quite popular right now. Rather not so innocently, prompts that generate slightly sultry pin-up art are also very popular.
The buyers on PromptBase are a mix of other AI prompt crafters looking to learn how to create better prompts (a lot of prompts are templates with tips etc), and individuals looking to buy prompts for a certain application.
Images or GPT-3 prompts?
Definitely image prompts right now, just because there’s more submitted. But GPT-3 prompts are still popular.
More sellers or buyers?
More buyers by quite a way.
Did you notice some interesting or unexpected patterns?
The popularity of models and the quality of images produced by different models is changing all the time. Right now Midjourney V4 is producing the highest quality, most versatile images I’d say. The popularity of DALL-E 2 has plummeted massively since it launched this Summer.
Now let’s go to the heart of the topic, Do you think prompts will be a huge thing in the next years and be more and more in demands as a new skill set or, will it fade away as the ai will become better at guessing user’s needs/ideas?
Some people have described a prompt engineer as “the final job on earth”. You construct a piece of text telling an AI model to do something, and it goes ahead and does it. No matter the industry: art, movies, medicine, agriculture; every task could be distilled down to a prompt.
Whilst I partially agree, I think it’s much more likely that “AI Communicator” will emerge as the dominant role in this space, with prompt engineering being an aspect of it. Text prompts are quite primitive, so I expect it will give way to more expressive mediums to achieve increasingly complex things with AI.
After reviewing thousands of prompts, there is definitely a clear difference between someone who can prompt well, and someone who can’t. I think prompt engineering will get easier, but as it weaves itself into more serious and complex applications instead of “fun ones” like generating images, it will require skilled AI communicators to work with the AI to achieve reliable outputs from it.
What do you think is the next big thing that gonna happen in the next couple of weeks/months, ai-tech related?
I think it’s a really interesting time right now where, over the past few months, developers have been given the biggest early Christmas present ever in the form of constant AI tech and models launching every other day.
It’s quite overwhelming the amount of low-hanging fruit industries this tech could be applied to; you almost don’t know where to start. I can imagine this is probably what it felt like during the dot-com boom.
As such, my prediction is that over the next few months we’re going to see an explosion in AI-related tools and startups appear, either from the developers quietly building things right now, or from those people who’ve been inspired to start building after seeing all this recent AI hype.
Subscribe to our newsletter!
What is the next thing you are going to try to build around Ai?
I’m fully focused on PromptBase right now. The mission has always been to enable skilled prompt engineers to earn from the time & effort that goes into crafting quality prompts. We just launched a service that lets you hire a prompt engineer for custom prompts, with users charging up to $300/job. In the works are some exciting new features to give prompt engineers even more revenue streams.
Where can people find out more about you?
You can follow me building PromptBase every day on my daily blog: daily.tinyprojects.dev