Ask a question
Introduction to New OpenAI Features for Bubble App Developers
If you're building a Bubble app that uses OpenAI, then this video is for you because OpenAI have just announced a shed load of exciting new features and I'm going to just break them down and explain how I think that they are going to be amazing for building AI apps with Bubble going forward. The date is the 6th of November, this is hot off the press and we're going to jump right in with the GPT models.
Improved Text Generation with GPT-4
These are of course how you can generate text within your application using AI. And first off we've got the fact that GPT 4 is now available in Turbo and the amount of text that you can send with your API calls has just gone through the roof. 128K. The example they give there is that's the equivalent of 300 pages of text in a single prompt. And not only that but it is also cheaper now as well.
The thing that we have to though consider as Bubble developers is that if you're using the Bubble API connector, there's still going to be that frustration with very long, large responses potentially timing out. So I'm really hopeful that Bubble is going to fix that at some point, give us much more freedom over how long an API call takes to time out.
OpenAI's JSON Mode and Reproducible Outputs
What else do we have here? Functions, I mean that was already there. Ah, JSON mode. So this is very exciting and I think JSON mode combined with reproducible outputs, this new seed parameter is going to take a lot of the trial and error out of trying to get that reliable and consistent formatting back.
I work with a lot of Bubble coaching clients and part of crafting that perfect prompt is so that you get back your response from OpenAI in that replicable and reliable structure in order for you to be able to pass, split, separate the data into your Bubble app. Well I think that has become a whole lot easier with JSON mode and reproducible outputs. The consistency is going to be so much easier and I'll be recording some videos on those in the coming days.
Upgrades and Improvements in OpenAI
Upgrades to GPT 3.5 Turbo, still the best offering out there for anyone who wants it cheap and who wants it quick. And so we've seen a number of improvements across that. The Assistant API, I need to dig into this a bit more, but this may be an appropriate way to handle many of the applications that I've seen people working with in Bubble that are bringing in OpenAI. So I think it's well worth digging into to see whether it is appropriate to apply to your Bubble app.
Basically what I'm taking away from this whole presentation, this whole press release, is that anyone who has used Bubble to build an OpenAI application or one pad for OpenAI, they can improve their Bubble app using many of the new features that are released today or releasing in the near future.
GPT-4 Turbo with Vision and Dory 3
We've got GPT 4 Turbo with Vision. Again, I'm going to be digging into the API documentation for that, seeing how easy or if possible it is to combine that. That would be like submitting an image of a puzzle or what's going on in this image to OpenAI through Bubble and getting a really accurate response. At least that's what we've been seeing or I've been seeing on posts on Reddit. Dory 3, again we've been getting sneak previews of this depending on where you are with your OpenAI account and we're getting some fantastic image generation there. I think finally we're drawing parallel with the likes of MidJourney, but leave a comment below if you disagree.
OpenAI's Text to Speech and Other Improvements
Oh yes, text to speech. I was only saying to a family member in the last week that we're now at a stage where text generation by AI can be superior than 9 out of 10 people in the room. That image generation by AI is getting close to indescernable photorealistic, undetectable photorealistic, but what is missing in the whole human mode of communication and understanding the world is the ability to take text and turn it into speech and for that to be convincing as a human. And I think that OpenAI have got very close, particularly the Onyx voice here which I'm going to play just to see if we get it.
As the golden sun dips below the horizon, casting long shadows across the tranquil meadow, the world seems to hush. And that is just a fraction below what I would expect to hear on an audible audiobook preview. It's so close. In fact, maybe if you were to play that over the phone with all the compression that goes into it, it would be undetectable as AI.
What else do we have? We've got lower prices, everyone loves lower prices, higher rates and then an improvement to Whisper. If you have built OpenAI into your Bubble app, then there is a good chance that you can save money and make improvements based on the features released today. And we will be recording videos and what we think are the most important. I think that the JSON mode in particular is going to be very high on our list. We will be releasing those videos in the next few days. So make sure that you like and subscribe on our YouTube channel. And if you've got any comments or any thoughts on anything I've shared, anything released by OpenAI, then please leave a comment. We love reading them all.