In this video we show using the Playgrounds feature to compare the responses of different models - such as Opus 4.6 and GPT 5.4 - along with different models and prompt settings.
Video transcript
Hi, I'm Alex, founder of Vile, and today I'm going to take you on a quick tour of our playgrounds feature. I feel like the best way to think about playgrounds is they make sense once you have prompts that you've maintaining for a period of time and we want to start to measure changes that we need to make to them. Perhaps we have revisions that somebody's asked for. We're trying to improve how they function or new models are coming out. We need to test the differences.
It's the best way to think about when you'll start to use playgrounds. We almost use them as a way to preview changes before we'll go through and investigate the logs on the prompt. To get started, what I'm going to do is press playgrounds here, create a new playground, and normally what I do is pre-select a prompt. Because I'm in a demo environment, I'm just going to write a really basic prompt here. I might say, give me one tennis tip for this topic.
And we'll put a variable in here like we would anywhere else inside of I know normally you'd be writing a much more sophisticated prompt than this. I'm just going to apply this. I'm going to say this runs on Opus 47. What I can do straight away inside the playground is just run these. And we need a topic here.
Why don't we say backhands and you'll see it's going to go through and it's now going to generate a response from Opus 47 on that prompt we built. But you know, this might be a real prompt and we're thinking, you know what, we need to change to GPT for different reasons or there's a new Opus model out. What I can do here is I could create a different column. What I'm going to do is clone this original one. And I might change this now to let's pick a different model here.
I don't think we need Pro here. Why don't we use GPT52 instant for example. I'm going to run this again with GP252 instance to start seeing the difference in these results. And this is a really powerful way as we're going to start through and maybe make some changes to a prompt, maybe make some changes to the model, the temperature setting, anything like that to start to evaluate how it's responding differently. One of the really things or one of the really things that I love about playgrounds, that was horrific English, but I'm going to keep it in the video.
We're one taking this is once I've got these results, I can actually press save on this playground here. And what's really nice is when I refresh, the results from my last run are here. And you can also share these playgrounds with other people on your team. This is something we use all the time. We have engineers on our consulting team that are working on iterating a live prompt with a customer.
or we're going to make some revisions. One of the first things we want to see is what did it look like before? What's the proposed change? And what are the other couple of things that we tried? And this is one of the easy ways that we visualize this before going into more details and perhaps doing a deeper test run.
You may notice that playgrounds also do support files. They support complex variables. You can have a large number of columns here. One of the other things we use playgrounds for sometimes is instead of testing seven versions of the same prompt, if we're sketching out a workflow in our brains, this is a nice way to visualize going from A to B to C across columns as well. So, as always, I hope this was helpful.
We'll start creating more videos and deeper videos based on the requests. This is a very quick course in how to use this playgrounds feature.