I had the call working but now it just won't work. I have it EXACTLY like you do { "model": "gpt-3.5-turbo", "messages": [ ] } but it keeps on giving me a: There was an issue setting up your call. Raw response for the API Status code 400 { "error": { "message": "None is not of type 'object' - 'messages.0'", "type": "invalid_request_error", "param": null, "code": null } }
Thanks! I’m in the process of creating an energy limiting chronic illness advocacy AI chatbot. The types of videos I’d be interested in are: tweaking personality and creativity levels for specialist chatbots eg requests for new campaigns turn up creativity temperature and questions about research get precise answers. Hosting chatbots for free/extremely cheaply for nonprofit/altruistic hobby projects (assuming Bubble chatbot needs to be hosted somewhere else? Or does it give you a public URL to share?). Understanding how to use different files in Bubble. Making the front end prettier. Using more than one API or plugins together.
Also from my initial chatbase chatbot I just have one long json data file which also includes persona information. Do I upload this as one file again in bubble or best to break it up into different files?
Hey Jenny, thanks for sharing with me what you'd be interested in. I will definetly take this into consideration when coming out with new videos! You can host bubble's applications directly on bubble. To be honest i'm not quite sure where else they could be hosted as it's part of bubble's pricing and business model to use their native solutions
Are you referring to fine-tuning a model/ using embeddings in accordance with your chatbot? Regardless I would suggest trying to remove or at least filter out any sensitive information in the json data
I would love to see a walk-through that actually starts with a button click, so the button will preload the prompt and ask the user to answer a few relevant questions for Assitant to then complete a response. Also, the initial prompt is referencing a record in a prompt repository (Type with Prompt Records). Also a UI with CHAT on LEFT and User Input on Right back and forth as a Chat. Thank you.
Hey there, thank you! You’d have to debug this to see if you’re even sending a proper request to OpenAI. I’d bet that something In your API call isn’t set up properly, if you’re using bubble use the step-by-step function
Great tutorial @Cerum Ai. i tried to build it.. it works. But how to do it for different users? As for now, different users will see the same thing & content
Hey thank you! Yea you’ll have to create the management of user accounts and display the completions based on the user that created them. We actually have a free template made specifically for this. You can check it out on site. cerum.ai/flow
Banger tutorial, thanks man! I'm trying to give my app some context by adding an initial prompt in the API Connector section. When I initialize it works fine but when I test the app it gives me an error saying that what was sent was not valid JSON. Here's what I'm putting in the API Connector: { "model": "gpt-3.5-turbo", "messages": [ {"role": "system", "content": "I want you to act as a copywriter and use the SOAP formula to generate sales content based on the input and refined queries that are entered"}, { "role": "user", "content": "" } ] } Any ideas what could be off here?
Hey Danny! From the body request you've show me in the API connector i can see that the dynamic element is just the "" part. Although you can do this it depends on how you have your workflow set up. In my tutorial you can see when I create a new chat (when Button ask is clicked) I create a new JSON Object. If you followed my workflow method this would result in an error as it's adding an object inside an object. I have the whole messages object set up like this { "model": "gpt-3.5-turbo", "messages": [ ] } I hope this makes more sense!
Hello again, hope all is well! I just have one error I came across and that is the max token error. Do you ever come across this and know how to reset the token count at a certain point in the conversation?
Thank you for this. I followed the exact format but I am failing to initialize call when I use the Chat or prompt in the api.. Could you share what you did there? Did you leave the user, role, prompt as is in the parameters?
Hey there! Can you refer to the timestamp you we're having issues with? and could you provide me more context regarding your error? I assume it's a JSON parsing error
ok yes to initialize it you need to with the setup I had in the API connector in 6:20 I added a sample message array (you can copy & paste this) {"role": "system", "content": "You are a helpful assistant."} Sorry I forgot to mention this.
This is great Dean. Thanks for this. Curious about the order of the rolled in the repeating group when you test it. The first question has the user’s question first (on top), and the second question has the response first and then the user’s question below it. How would we set it up so it goes question, answer, question, answer like it does in cbatgpt? Thanks in advance.
Hey Danny! Of course, and yea I noticed that in the tutorial the order of the repeating groups isn't in a traditional "customer & assistant" conversation that we see with ChatGPT. I'm sure this can be an easy fix, I'll look into this very soon here tomorrow and get back to you on this! Also feel free to join our newly created discord of other no-code devs and ask other questions you may have (i'm more responsive in there)
Theirs a few ways to do this. The most optimized way would be with using embeddings with similarity search as it can handle large amounts of text data while being cost effective too. We’re actually developing a no-code solution for this specific reason. Stay tuned!
Hey! Awesome tutorial. This is a Univeral bot and if I ship this it's gonna be overlapping with different users and is basically gonna be unusable- Are there any way to make this user-specific?
Hey there Malte, thank you! And yes I'm very glad you noticed! You would have to tie each conversation to say something like a conversationId (unique ID generated) which is dependent on the user that's signed in & created the conversation. So it would only show the signed in users relevant conversations I hope this helps a little. I'm also a lot more responsive in our discord if you have any other questions. The link is in bio for that
Very helpful, thanks a lot! Do you know if there's a way to start the conversation with a system prompt without it showing up in the messages displayed on the screen? I want to tell my chatbot to answer in a certain way and I think you can do that by sending it a message first, but I don't know how to set that up
Hey there! Have you tried initializing it with a system message prompt in the API connector? This would be an object within the array. Such as this {"role": "system", "content": "You are a helpful assistant."}
Hey Alivio, to create a login/signup portal you would have to create a separate landing page for this and set up the workflows to create an account for the user. From there you would direct them to the actual chatbot application. If this is something that would interest you guys I could make a video on it!
Hey there glad you enjoyed this tutorial! And do you mean automatically scroll down as messages are entered into the conversation? Like how a traditional message system works? I'd have to look into this, may need some custom embedding to offer this UX functionality
@@cerumai Yep exactly. Right now, the repeating group fills up and I have to manually scroll up to see the result for every message. I thought there could be a simple feature to get it to work but I am afraid it's not straight forward. By the way, it's just related to Bubble and not GPT.
@@brianelley1045 of course yea I’ll look into this and get back to you. I don’t imagine it’d be complex at all! Make sure to checkout our newly created discord as well if you want to connect with other no-code devs (I’m more active in there as well)
Followup question: My repeating group shows user throughout, instead of user/assistant and I obviously don't get a response. I followed the exact same formula. Something seems off... any idea how to fix this?
Hey Thembi, the display element's are also found in the workflow. In my tutorial, in workflow steps 1 & 3 you can see the "Text" field we're creating in our database which consists of the users input value & the result of the API we get in step 2. In our design editor make sure you have the data source of Chats in the repeating group and you display the "current cell's Chat's GPT role's Display" & " current cell's Chat's Text" you can check in the database to make sure in the chat database that these are being chat messages are being logged. I hope this help
Hey John, the dynamic JSON value would be provided by the user along with the workflow that we've in set up in the backend to feed this. If you're getting a JSON error I would try reinitializing the API call in the backend (at 6:27) with an example JSON object and saving it. Make sure "allow blank" is also checked. You can try plugging this into the value parameter to reinitialize it {"role": "system", "content": "You are a helpful assistant."} Let me know if that works!
