Building with ChatGPT - Part 6 - Prompt Engineering Thoughts
A look into the future of prompt engineering & what's coming up next.
In this weeks blog post, I’m sharing my thoughts about the value of prompt engineering, what is happening to RecapIt and the Startup AI Advisor and what I’ll be building next (preview: An AI tutorial and an experiment for autonomous DAO governance).
On Prompt Engineering
Prompt engineering has certainly taken the Twitter AI world by storm. If you’re following people in the AI field, your “For you” page will inevitably contain a few Twitter threads like “500 Marketing ChatGPT prompts you need to know”. I think that these are mostly not worth reading, because no one will remember anything from that many prompts anyway. And quite frankly: Most of these prompts have probably been generated by ChatGPT itself, because
a) some of them make barely any sense
b) no one had the time and experience yet to seriously create 500 quality prompts by hand.
Don’t get me wrong: I think prompt engineering has some value. But you can achieve a lot with some basic techniques already without needing hundreds of prompts.
There’s also an ongoing debate about whether prompt engineering is just a fad or whether it will be a job that every company requires. I think short-to-medium term understanding promting techniques well certainly is a big advantage. At the same time I believe that in the long term AI systems will become advanced enough to just do the heavy lifting by itself: If I ask ChatGPT a marketing question, I shouldn’t need to tell it beforehand that it is a marketing expert or to make sure to ask me questions about things that are unclear.
TL;DR: Prompting techniques are useful now and in the medium term. Getting 500 ChatGPT prompts for a topic is worthless. Long term, AI will do the hard work by itself and there’ll be no need for prompt engineers anymore.
No more RecapIt and Startup AI Advisor
If you’ve been following along, you know that we’ve been building a YouTube video summarization and chat bot using Langchain and Streamlit. At the time of writing over 300 unique videos were already analyzed 🤯 However, I won’t be updating the app in it’s current form any further going forwards, because:
- There’s already a few similar apps out there, also as ChatGPT plugins. It’s kind of a low hanging fruit.
- I expect monetizing will be difficult, especially with the low-ish amount of traction it got. The only possible way towards revenue would probably be by offering the service to YouTube creators and their audiences (with a few upgrades like instant summaries and channel wide Q&A).
I’m happy about having launched RecapIt though, because it gave me some valuable insights into coding with Langchain which I can utilize going forwards. If you’d like to look or use the code that was used to build the tool, check out the previous blog posts in this series:
David
Similarly I won’t continue work on the Startup AI advisor. A few weeks ago I thought that the idea was good, but by now my thoughts on prompt engineering have evolved. The envisioned advisor app was mostly about prompt engineering and it’s not worth putting in a lot of energy into something that will become obsolete in some time not too far away. In addition, there’s several tools out there already that allow you to “create personas”. If you need some inspirations for prompts to use for these tools, have a look at this blog post:
David
What's Next
I'm currently working on a tutorial for solo-entrepreneurs and how AI can supercharge their business. It will focus on practical tips and also have an AI tutor that has been trained on the study materials. Keep an eye out for the official launch!
Besides that, I am going to build an experimental AI voing bot for blockchain DAOs. The idea is to train AI on the projects documents, let the user set some voting preferences and then ask the AI to vote on governance proposals. This hopefully will help in solving the "no one is participating in governance" problem that DAOs currently face. Next weeks blog post will outline more details on this :)