everything literally everything I am searching since 1 month. please make more video on this topic for complex situations. please make video I really need it. thanks a lot
Highly useful! In conditions when we have huge amount of marketing AI videos, and small amount technical ones, your video is very interesting and efficient! Thank you!
Bro imma be honest with you… i have seen a lot of videos trying to to understand how to integrate custom functions in my assistant, but none of them worked for me until i saw your video🙏🙏 thanks a lot, you have great teaching skills 🤙 u r the best
really nice video, explaining things very well. The thing i like the best is that Sam talks at normal speed, and explains everything very simple with clear examples. Thnx
how can i call function just like you? In common, we add message in thread, but this video send instruction to run. help me. and your notebook link is not collect (other file)
Thank you for the tutorial, it was very much appreciated, and very well explained such that even an amateur such as myself can understand it. I will be building an assistant to learn about my work site, as a veryquick example -- showing it a diagram of the door closing mechanism on the front entrance door, so that in the future it can help with troubleshooting/advising how to make a certain adjustments such as the door closing speed. Over time I'll have something similar to the computer on the Enterprise.... maybe. Anyway, it might be a fun video for you to consider making, and selfishly I might also get some pointers or ideas when seeing how you approach such a project.
This was very informative. I’ve thought about doing something like this to take my current location and maybe a compass direction and use an adsb API to fetch the planes in that direction from me and return them in order of how close they are to me. Additional feature might be to use a plane reference site to gather additional details or emphasize more unusual ones. This would let me essentially ask “what’s flying low to the west of me” and have it describe the sub hunter plane that sometimes trains around here. It’s essentially the same thing as me pulling up a flight radar app on my phone, then googling for details about the one that caught my attention. If practical, this might be an interesting follow up.
Excellent video, it was very helpful. Could you attach the notebook that you show in the video in order to make the interpretation and understanding of the concepts easier?
I am newbie in coding so already sry for the dumb question but... where is this function deployed? Its running on your computer? For my understanding the information that we provide to the assistant is only what input are needed but the exact logic is not stored under "functions"? What if I dont want to run the logic on my computer? Do I have to deploy the function somewhere and put an API call as function for the assistant?
Thank you very much. Could you please share the ipynb file you used in the video? The file mentioned in the comments seems to be different from the one in the video.
I had a bit of a dig into this - it looks like some people have been experimenting with this on the OpenAI forums, and it looks like someone has found a way of doing it via Retrieval. Haven't tested it, but might be worth a try! community.openai.com/t/functions-in-the-assistant-api-playground/479721/8
Thank you for the video. The notebook you had shared is openai_assistant_retrieval.ipynb. Are you able to share the notebook openai_assistant_functions.ipynb you used in the video? Thank you
i am creating that chat bot but i got to know that even in threads model didn't remember the previous chat for example i wrote a prompt "change weight in previous prompt and keep rest details same". will that prompt work?
I think as long as you're working in the same thread (and your context/conversation isn't too long), the assistant will reference that thread in the response. Hope that helps!
The bot answers the question based on the data that was uploaded via the aviationstack API. How can I make the bot itself upload the necessary data? So that it doesn't need to enter the flight number in the aviationstack API.
Hey - so two approaches here, one could be to fetch data from your own database (so instead of using AviationStack, you’d query your own database), or you could potentially use RAG (or Retrieval) to fetch context from provided documents, if the information is static. I have another video on OpenAI Assistants Retrieval which might help here. Best of luck!
At the step with assistant = client.beta.assistants.retrieve( assistant_id=assistant_id ) assistant I have the error: timeout: Override the client-level default timeout for this request, in seconds """ extra_headers = {"OpenAI-Beta": "assistants=v1", **(extra_headers or {})} --> return self._get( f"/assistants/{assistant_id}", options=make_request_options( extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout ), cast_to=Assistant, ) I don't know what to do(
The part I don't understand yet is, with the Python code where do you host it so the Assistant can call it. Then in the OpenAI assistants dashboard under functions say you have function Name: Get_weather and you hit test button. How does the Assistant know where the code is in relation to Get_weather? There is no endpoint etc set up? Im so confused on this part. Someone please help! haha
Hey! Kind of a late response but still. The model doesn't really call the function, so it does not need to know where the Python function is. All the model does (in a general sense) is taking the question/message from the user, taking a bunch of **descriptions** of some functions (and their parameters), and returning the name of the function to call. Basically the model only **tells** you what function you should call (and what parameters you should call it with) based on the descriptions you provided. After you get that response from the model, is up to you to call the function the model suggested (as stated in minute 8:30). And after doing this, you can optionally pass the values returned by these functions back to the model, so the model can come up with a more "human sounding" response (which is what's happening at minute 10:42). Hope that helps!
