Getting my hands dirty with AI prompt engineering
Was fun a week as I got to work with our developers on making some AI prompt engineering tweaks.
It’s been a fun week in developing more of our AI.
I know, I know, AI seems like a flashy buzzword these days and everyone and their mother is building some AI tool or custom GPT. But I figured I’d talk a bit about the most interesting part of building our custom AI chatbot.
As most of you know by now I’m the cofounder of an immigration tech startup and we have an AI chatbot that helps immigration law firms save time with research, writing, etc.
As the resident non-coder it’s been an interesting experience helping to test and tweak the chatbot, although most of what I help with is the prompt engineering.
For those who haven’t read the term, it’s not as scary as it sounds. Think of it like this:
When you’re building custom AI, you’re giving it extra details and instructions that your user can’t see. For example, when you open up ChatGPT and type in a question, there is a LOT more going on in the background than we know.
This is because there are prompts that OpenAI have built into their tool to direct it in how to find information, where to search, how to search, how to summarize, how to repurpose the text and send it back to you.
Same thing for most of the custom AI tools out there.
As another example, we were re-working one of our key prompts this week, which helps us determine the type of question a user is asking.
Let’s say one of our users, usually either an immigration lawyer, consultant or their staff, log into Visto to use our Copilot. They might ask something like “what documents are required to apply for a work permit in Canada?”
In our case, one of the first steps is for the AI to determine what type of question this is: is it a legal research question, a drafting question, maybe something else entirely.
So we have to come up with a prompt to help it determine what type of question it is, then where to go to find the answer, then how to structure that answer, etc.
Now for most of you this might sound super technical and super lame, but I’ll tell you that, at least for me, it’s really fun and does not require technical skills, per se.
I’m not a coder, and while you do need some very basic understanding of how the AI and your specific tool work, what’s most important is being able to write prompts in clear, concise, precise language.
And as a lawyer and big reader who is quite the fan of word-smithing I find it quite the challenge, especially since there’s no such thing as “perfect”.
It’s about tweaking, testing, re-tweaking, re-testing, until the output from the AI is the quality that you’re looking for. In that sense it’s also a new kind of challenge, not just something you can look up on google or pull from chatgpt and be done with (although, ironically, AI can be quite helpful with coming up with prompts…).
So that’s some of what I spent my time on last week that I really enjoyed.
Not only do I enjoy the practice of prompt engineering, but as AI and the tools themselves become more powerful, prompts and the design behind them will only become more valuable and important.
Have you had a chance to work with AI prompts? Any tips or advice?
This was definitely one of my more technical posts as I normally focus more on sales, marketing, operations, etc., but this AI stuff is super cool and surely not going anywhere.
So hopefully you enjoyed, and maybe even want more AI details and updates in the future? Let me know.
Have a great day,
- Josh Schachnow
Canadian immigration lawyer, CEO at Visto.ai
If you’re looking for more, definitely check out: