Building with ChatGPT - Part 1 - Prompt Experimentation
Back in February, when the ChatGPT hype was in full swing, I started experimenting with it along with everyone else.
I was just starting to wind down my work on my startup and reflecting on what went right and what went wrong - amongst the latter probably was the fact that we hired too many people that we probably didn’t need too quickly. So then my idea simply was to see how ChatGPT would perform in certain roles - like marketing. My expectation was that it wouldn’t necessarily outperform the humans we hired, but that it would “win” in a comparison considering the price vs reward tradeoff. I started experimenting with prompts and my expectation was correct. ChatGPT gave me answers that were “good enough”. I decided to tinker with the idea a bit more and build a small application for solo startup founders: The plan is to give the user access to different “personas” - like a marketing or a technology officer - that act as a sort of “AI advisory board”. I initially wanted to just build the app out in Python since that’s a coding language that I’m familiar with, but I quickly realised that I could tinker with ChatGPT even more by letting it help me write the code in node.js.
This personal-blog series is a documentation of the journey: Experimenting with prompts, reflecting on how the app would be helpful in startups, building an app with ChatGPT and following along whatever new developments are happening in the AI space. The eventual goal is to release the app to everyone free of charge, but that is more of a “side artefact” - the real value lies in the road to get to that point. This is more of a side project, so expect around 1 short-ish article each week.
Why ChatGPT
I’ll be using the field of marketing as the main example here. Every startup needs to do some marketing and community engagement - but does it really warrant a full-time position with the associated expense? Or even a part-time position? Is it worth onboarding a person and making sure they are always up-to-date and on the same page as you as a founder? If you are early, then ChatGPT may not necessarily do better than a person, but it will most likely perform better from a ROI perspective (considering the salary costs plus the overhead organisational costs that an extra person brings). ChatGPT is perfectly capable of writing copytext for your landing page or a PR release about a new feature. As for community engagement - you can make use of some chat bots in that area as well, but if you’re early it’s probably better anyway if you as a founder keep in touch with your customers.
Experimenting with Prompts
Ok, so we are relying on ChatGPT for our marketing. Let’s pretend that you as a solo founder know next to nothing about marketing. We can ask ChatGPT to give us an intro:
That’s not too bad - it contains a few good points like identifying the target audience - but we can definitely do better by providing ChatGPT with some context:
- We know nothing about marketing, so let’s make ChatGPT explain it like we are 5 years old.
- Our startup is a tech B2C product. We should probably let ChatGPT know.
- Another good trick is to tell ChatGPT to act like a certain persona - say an experienced marketing manager who has taken startups from 0 to a whole lot of revenue.
Let’s see what results this gives us:
That’s a lot easier to digest and gives us some concrete actions (like creating Social media accounts and tracking our efforts)! We could now keep asking for details on specific steps, but instead let’s get down to a specific task (while staying in the same conversation so that ChatGPT doesn’t loose the context):
That’s a good starting point. With some tweaking we can probably use that. Let’s see how we can create the landing page ourselves:
And just like that we’re ready to go! Obviously nothing that ChatGPT does will be perfect, but considering that it is free and cheap, the ROI and productivity increase is incredible.
Here are some more prompts you can use in your startup - also for non-marketing positions:
David
Why an App
Initially I thought it would be useful to just keep a sort of “prompt book” somewhere on my computer and just copy-paste the prompts into the OpenAI online chat interface. But I quickly realized just how inconvenient that was and how I sometimes refrained from using ChatGPT because of the additional friction. So I decided I’d just make a simple app where I could choose the “advisor type” and have it remember the description of the startup idea(s).
My background is in Python development, but I never really went into frontend development or “web stuff”. So my plan was to just code up a simple Python app. Once I was 30 minutes into playing with the OpenAI API I however realized that I could just play around with ChatGPT and make it help me make a node.js app instead! That way I could see how well ChatGPT works for coding tasks and I’d get a nicer UI/UX (hopefully) than what I would get with Python.
Next weeks blog will be about how the ChatGPT-assisted coding journey started.
