Hey everyone! I wanted to suggest sharing some tips for asking great questions in ChatGPT or Bing Chat for Bubble.io related questions. It seems that ChatGPT and Bing does not have enough knowledge about Bubble, but maybe there are better ways to phase it?
If you’ve had success getting answers to your prompt, please share your tips! Maybe you have a certain way of phrasing your question that always gets a response, or maybe you have some tips for providing enough context so that AI can understand your question better.
Let’s help each other out by sharing our tips for asking great questions, so that we can all get the answers we need!
For example, I start from
"I want you to act as bubble.io expert. I will ask no-code programming-related questions and I want you to guide me on how to {your question} steps by steps. "
GPT-4 (only available via API or in “ChatGPT Plus” ($20/month at present)) does know some things about Bubble (such as this) and can provide fairly reliable information about Bubble. Prompting GPT-4 to be an expert in Bubble won’t actually improve things, but you can set the context of your conversation like, “I have some questions about the Bubble (bubble.io) visual web development platform. [Question 1…]” and that’s a helpful introduction.
(But telling these things to “be an expert” in such-and-such does not make them better at that particular subject. There’s a lot of folk wisdom nonsense around prompts – both in large language models and in text-to-image models – that isn’t actually helpful. I call these sorts of things “totemic prompts”. They might make you feel good, but they don’t actually do anything meaningful. GPT-4 doesn’t need that sort of thing anyway. If it provides an answer that’s not clear, you can simply say, e.g., “I’m a novice here, can you provide more detail?” or “Why do I have to do step 5?” (GPT-4 loves to put instruction steps in numbered lists!))
It’s important to note that GPT-4 is radically more knowledgable about… well… everything, including Bubble. Earlier versions of GPT (e.g., basic ChatGPT / GPT-3.5) do not know a significant amount of information about Bubble and will respond to all questions with misleading information. (That is, the answer is “hallucinated”/authoritative-sounding bullshit. In this case, you’re essentially talking to GPT about a fictional product and its answers are fictional as well.) While it can be fun to talk to those earlier models about “Bubble” literally none of the technical information they might provide is accurate.
GPT-4 doesn’t know a deep amount of info about Bubble, but can probably answer most basic questions and could propose interesting solutions to more macro questions about, for example, methodology one might follow to implement a shopping cart or whatever.
Because it’s basically undocumented, GPT-4 knows little about plugin development for Bubble, but I’ve been working on a rather long “build Bubble plugins” prompt and I talk about how I’ve been creating that and give a sneak peak at how it works in a video posted in this topic:
Not really needing GPT-4 to be my helpful pro guide to doing basic stuff with Bubble, I haven’t probed its knowledge of vanilla Bubble deeply. You might find it to be very helpful. (I would keep in mind that specifics of the Bubble interface may be somewhat unclear to it as it hasn’t literally “seen” them and the amount of Bubble-related content it’s ingested is many orders of magnitude less than for code-based web dev tasks. It might call something by the wrong name, but it’s close enough for jazz.)
I don’t want to unersell its capabilities with respect to Bubble. They might be really good on the basic stuff. (If you follow that link at the top of the post, you’ll see it knows quite well that you can’t just add 1 to every item in a numeric list without jumping through hoops!). Also, it knows about the more popular and well-discussed plugins, including List Shifter. And it still blows me away that it knows that I’m the author of that plugin. Spoopy!
Hi Keith, thank you for sharing your insights, especially regarding GPT-4. Since not everyone has access to it yet, it’s great that you have been able to experiment with it and share your learnings. I’ll keep an eye on your post.
As a new Bubble developer, I have played around with GPT3.5 and found that it works well for basic tasks at the macro level. However, it often starts talking about fictional products, which limits its usefulness in developing app as your cop pilot.
By the way, I have found using Bing Chat to find related forum or YouTube videos helpful.
