The Power of OpenAI API in Bubble Apps
The OpenAI API can add the full power of ChatGPT to your Bubble apps. You can use their API to add text generation within your Bubble app and do a whole host of amazing and cool features. For most people they will be choosing between whether to use GPT 4 or GPT 3.5 Turbo. In this video I'm going to compare the two and offer some points of some comparison and also run a few tests in a quick Bubble app mock-up that I've created.
Cost Comparison: GPT 3.5 Turbo vs GPT 4
To begin with I'd say the main difference for many people is going to be cost. That we have a significantly lower token cost on the GPT 3.5 Turbo models compared to GPT 4 which is up here. So that says if I've got my decimal place in the right place double the double the cost. Yeah I got that right maths is fine. What is a token? Well a token is explained in the paragraph up here. So this paragraph is 35 tokens.
Choosing the Right Model for Your App
If you are using and building an application where you want to let your users kind of run free, obviously put some constraints with ChatGPT and OpenAI's API, then I would recommend using GPT 3.5 Turbo because it is going to cost you significantly less compared to if you had some sort of power user or you haven't put limits on your system and they're using GPT 4. The reason for the price difference is because first of all GPT 3.5 Turbo is trained on a smaller model. It's still huge that's how we get the amazing text that we do from it. It's also a much older model it's been around a bit longer. GPT 4 is trained on a large model and it is a newer model hence the increase in the numbering system.
Comparing Response Times and Quality
Let me take you through to this page and I'll just refresh it. So I'm going to say write five tweets promoting a YouTube video about comparing GPT 3.5 and GPT 4. Copy that into each one and clicking the buttons only a moment apart because I think the first thing we'll notice is of course that GPT 3.5 Turbo is quicker and let's wait for that to come through. We should see that any moment now. Yep there we go there are tweets from GPT 3.5 Turbo and we're still waiting on GPT 4.
Considerations for User Experience
So there's two things to consider here at least one is the weight that you put your users through I'd really recommend if you are not using a plugin that allows you to stream the data in if you're using the API connector that you add in some sort of loading animation and we've got some videos on adding loading animations to OpenAI on our channel and on our website. The other thing being that if you're waiting for a big response or you put a large prompt in with GPT 4 you're still you still got the chance of running into an issue with it timing out with Bubble. Bubble has increased the time that it will wait for a response from the API but GPT 4 can still take a significant amount of time.
Comparing Output Quality
I wonder if there's just anything that we can notice to compare here so they've both got emojis in them. I don't know if it shows particularly clearly here my experience with using GPT 4 is that if I want a real richness of text here just to take up to the next level with it appearing like it's written by a real human being I think that GPT 4 does edge ahead so there are indeed cases in my daily workflow where I would choose to use GPT 4 but then again on the other hand I'm happy to have a much quicker response and something that isn't isn't so rich in the use of its English of the words that it places in it and I will use GPT 3.5 Turbo.
Testing Longer Form Content Generation
Let's just try a longer form why don't we go with write a 500 word blog post about the pros and cons of using Bubble.io to build a web app. So again we'll expect GPT 3.5 Turbo to come back quicker and we're going to wait and see what happens. Okay so we have a much longer wait on GPT 4 remember we don't want it to time out that's really going to frustrate users if you put that in your Bubble app but I wonder if we can see much of a difference now I wouldn't say that the fact that it's put a header in here and it's put numbering I was running some tests as a preparation for this video and GPT 3.5 in that case with a very similar prompt also laid it out like that.
Subtle Differences in Output Quality
I would say though an expression like Bubble.io boon or bane for web app builders you might you must personal choice but maybe that is going into that region of a bit more of our richness of a creative writing approach. We got a slightly longer response but again that's not really difference between the two models I mean basically in a nutshell it comes down to cost weight and I would say if you're writing long-form content then I have started to use GPT 4 instead of 3.5 in my daily workflow because I just want that little bit extra more whereas yeah like I say there's many occasions when I will use GPT 3.5 Turbo instead.
Conclusion and Feedback Request
I'd love to hear your thoughts though please leave a comment down below if you are using GPT 3.5 Turbo or GPT 4 in your app and what has nudged you one way or the other I would love to hear your feedback and we read every single comment.