Legendary video Mark! I was a bit overwhelmed by the api documentation in terms of how broad it is. This video absolutely helped me in gaining that necessary overview to design my applications software architecture. I love how you just show a working example with all components loosely integrated to demonstrate how all of the components work together. You probably saved me a day of work going back on early uninformed choices in a 30 min video. Subscribed!
Hi Mark. Thanks a lot for this great video. I played around some months ago with the v1 of the assistants API. With this background your video helped me a lot to understand what has changed. I liked the idea of using Perplexity API to get information about recent information which is not part of OpenAIs huge LLMs.
Honest, Straight to the point, insightful, that's what we like, good job mate ! By the way i'm curious about the scrapping assistant for documentation, do you have any tutorial about this ?
My pleasure! Appreciate your appreciation I use FireCrawl to help me with a lot of documentation scraping - if it misses anything I get my VA army to help me with the rest
this was very helpful, Thank you so much. Can you also dive into the portion where you define the actions in custom GPT to talk to replit microservice? (The OpenAPI schema you used to make that connection)
I have a problem that when I upload files in the search tool, some answers work to answer from outside these files. Are there specific instructions for it?
That would be a prompt engineering component - make sure a certain set of criteria or threshold needs to be met to trigger the tool. I would designate a whole portion of the prompt for this.
Hey Mark, so I was making the assistant of the ChatGPT website itself but when I run it with my large documents that are in the vector search, it reaches a token limit and I wont be able to ask any more questions until I clear or restart the assistant. Im confused. Using Chat GPT is allows for long conversations. Doing the process you explained, would this allow me to better communicate with the assistant without it hitting a wall? Maybe Im going about this wrong or not being as efficient.
Hey Rob! The assistants API has a per document limit of around 5 million tokens; I notice that even at the 3-4 million mark, the indexing slows down quite a bit. Because of that, I convert all PDFs into txt files, then split those text files using a text file splitter online. So if I have a document with 5,000,000 tokens, I’ll break it up into 1,500,000 token documents so that it can properly index them.
Say Mark would you be so kind as to update with the link to the documentation scrape? That would be so useful. Or if you could let us know what you used to scrape the documentation, that would work too. Thanks!
Hi Joe! RUclips doesn't let me attach Google Doc links, but I managed to get around it by adding bitly links: 1) Part 1: bit.ly/3VDXC4Y 2) Part 2: bit.ly/3xeFCVz I had my VA do this, but if you wanted to try to do it in an automated fashion, I would recommend trying FireCrawl and entering the URL. Will be making a video on this in a few weeks.
Legendary video Mark! I was a bit overwhelmed by the api documentation in terms of how broad it is. This video absolutely helped me in gaining that necessary overview to design my applications software architecture. I love how you just show a working example with all components loosely integrated to demonstrate how all of the components work together. You probably saved me a day of work going back on early uninformed choices in a 30 min video. Subscribed!
So so happy I was able to save you some time!! Thanks for the sub 🦾
Man, I love this deep dive for a newb sub, visual/audio learner 55+ old fella. Beers on me when you are Aotearoa Chur Koko...🎬🦅🙏 bring LO...
Thanks ! Great and concise, keep up the good work
Thanks so much for the kind words!
Hi Mark. Thanks a lot for this great video. I played around some months ago with the v1 of the assistants API. With this background your video helped me a lot to understand what has changed. I liked the idea of using Perplexity API to get information about recent information which is not part of OpenAIs huge LLMs.
I see you binging my content haha!
So happy you’re finding value in it 🦾
This was an exceptional video. Thanks for that.
thanks Aaron! Glad it was helpful 🦾
Honest, Straight to the point, insightful, that's what we like, good job mate ! By the way i'm curious about the scrapping assistant for documentation, do you have any tutorial about this ?
My pleasure! Appreciate your appreciation
I use FireCrawl to help me with a lot of documentation scraping - if it misses anything I get my VA army to help me with the rest
this was very helpful, Thank you so much. Can you also dive into the portion where you define the actions in custom GPT to talk to replit microservice? (The OpenAPI schema you used to make that connection)
I have a problem that when I upload files in the search tool, some answers work to answer from outside these files. Are there specific instructions for it?
This is a known flaw for the assistants API unfortunately- the best you can do is in your prompt, to tell it to refer to ‘file names’
keep update more video, it's so good. Thank you
Thank you so much for the feedback! I’ll keep the content coming for sure 🤙🏽
THe link to the scrape of the api documents is wrong it's to your website instead.
thank you for pointing out! just fixed
you can just add the "file_search" in the tools array when you create the assistant
Thank you! I can’t believe I figured everything else out and not this.
Much appreciated 🙏🏻
this is super helpful thanks mark
my pleasure Samin!!
During the chatbot conversation, how to ensure that the tool call its used only during a specific step, and not for all the user queries?
That would be a prompt engineering component - make sure a certain set of criteria or threshold needs to be met to trigger the tool.
I would designate a whole portion of the prompt for this.
Hey Mark, so I was making the assistant of the ChatGPT website itself but when I run it with my large documents that are in the vector search, it reaches a token limit and I wont be able to ask any more questions until I clear or restart the assistant. Im confused. Using Chat GPT is allows for long conversations. Doing the process you explained, would this allow me to better communicate with the assistant without it hitting a wall? Maybe Im going about this wrong or not being as efficient.
Hey Rob!
The assistants API has a per document limit of around 5 million tokens; I notice that even at the 3-4 million mark, the indexing slows down quite a bit.
Because of that, I convert all PDFs into txt files, then split those text files using a text file splitter online.
So if I have a document with 5,000,000 tokens, I’ll break it up into 1,500,000 token documents so that it can properly index them.
Both of your colab links are the same, can you post the one for the function calling?
thanks for pointing that out @Trevor!
just added it :)
Heat every time you drop Mark🔥🔥🔥
Appreciate you brother!
Say Mark would you be so kind as to update with the link to the documentation scrape? That would be so useful. Or if you could let us know what you used to scrape the documentation, that would work too. Thanks!
Hi Joe! RUclips doesn't let me attach Google Doc links, but I managed to get around it by adding bitly links:
1) Part 1: bit.ly/3VDXC4Y
2) Part 2: bit.ly/3xeFCVz
I had my VA do this, but if you wanted to try to do it in an automated fashion, I would recommend trying FireCrawl and entering the URL. Will be making a video on this in a few weeks.
@@Mark_Kashef Thank you so much! Really appreciate it!
@@JoeKnouse pleasure!
This is amazing. By the way Mark, the link to replit seems broken
Thanks for pointing out! It was accidentally set to private, you should be able to see it now :)
Please add time stamps
Thank you for the feedback, will do :)