The Boolean Generator is a powerful tool on Pulsar to expeditiously craft Boolean suited towards the topic of your choosing. However, it can take some practice to get exactly what you're looking for out of the tool. This is due to the non-deterministic nature of the LLM used to power our Boolean Generator. Have no fear; in this guide we'll give you 4 tips to make your experience with the Boolean Generator as smooth as possible!
1a. Give the tool examples.
Sometimes repeating yourself isn't such a bad thing. When running a prompt, consider using language that repeats the point of your query in different ways. For example, if I was looking for ice cream flavors. I might start by looking just for "ice cream flavors"
Not quite what we want, is it? But if I add some examples:
It's getting a bit better! It's mostly just used our examples here but it's a good start.
1b. Give it BAD examples too!
Bad examples can help. Let's add a bad example to get rid of generic mentions to our ice cream flavors search.
Excellent, this got rid of our generic term. It may have added a blacklist we word we don't necessarily need, but we can always delete; we want more from this to start!
2. Emphasize what you want with repetition.
We have our examples, but it's not elaborating. We want more. Let's ask the tool for more flavors by repeating what we're really looking for.
Awesome! The blacklist is a bit of an issue, but the main body is looking much better. Our examples have extrapolated into more terms for our search. So what's next?
3. Exaggerate!
LLMs are all about the dramatic. Ask for more! In our ice cream example we may be able to even cut back on the repitition by being dramatic. Like so:
WOW! Now I am absolutely sure that's not every ice cream flavor ever made, but that's a much bigger list than we got before, and we said even less! Notice we are still repeating ourselves, but the exaggeration helped the system understand we mean business! So, our final tip:
4. Change a word or two and try again.
As we said at the beginning, LLMs are non-deterministic; this means even the exact same input won't necessarily give the same answer twice. However, we can use this to our advantage; as it also means sometimes slight changes in input can cause more major changes in output
Let's just try one last change. I like our boolean above but maybe I think it's got a bit much. What happens if I take one example off?
And it's still really good! Just a little shorter, and a few less exclusions (thank goodness, I really wanted to see stuff about ice cream trucks!)
A Few More Examples
Here are a few more examples of good generations. The first one is someone looking for mentions around a brand and its products. The second is for someone looking around traditions that might be common around the Thanksgiving holiday.
The system will change and improve, so some of these tips may become obsolete over time, and there may be even better tips out there! Remember that you aren't stuck with what it makes; you can edit the Boolean to your heart's content!
Find any cool tricks to get the best Boolean? Or just want to share what you made? Reach out to us on intercom with whatever you got! The more feedback we get, the better this tool becomes.