Thank you Bo! I built an agency a few months back now and I have been trying to get down all the theory and groundwork done to have my building blocks in order. I watched Liam's video on understanding chatbots as well as prompt engineering but quickly realized that GPT's came into play changing the way we as agencies build out and utilize these tools. This is exactly the video I was looking for! I aim to start my UA-cam journey here in the near future but I want to understand all of these key concepts while documenting my learning journey! I only came across your channel after the group podcast you guys did that was insanely helpful!
So you started agency and now thinking about starting UA-cam for personal branding, isn't it better to start UA-cam and run your Agency as you get clients from UA-cam
@@NewsSketches Yes and no... you don't need UA-cam to get clients especially in the automation world there's thousands upon thousands of people looking for solutions we can provide. X and Linkedin are 2 great platforms to find clients I've mostly just been networking and automated cold outreach messages to expand my reach hands off.. I mainly want to start my UA-cam channel to document my own journey as well as help people who want to work in the automation field if I get clients along the way from UA-cam then that's a bonus but I think staying true to yourself in and of itself is the core concept of developing a personal brand, if I'm just in it for the money then I won't enjoy it.
@@bo-sar Hey Bo, I was confused about one part. From my understanding, the language model receives the user's query as a text input and then tokenizes it into individual words or subwords. Chunking, from my knowledge, is a manual operation. I originally had the misconception that 4 letters equaled 1 token - I forget where exactly I heard this, but that's not the case; 4 characters is just an average approximation of the length of a token in English text. Even longer, common words in the dictionary are still likely to be handled as 1 token regardless of the word's length. For example, words like "Mississippi" or even "Supercalifragilisticexpialidocious" are most likely treated as 1 token in a large language model (LLM) like GPT-4o. It's only when we encounter rare or made-up words that aren't in the model's vocabulary (like "Flibberflabber") that they might be split into multiple tokens. Additionally, words like "mother-in-law" would be treated as multiple tokens because they include a hyphen and are considered separate subwords. So, chunking only needs to occur when we exceed the LLM's token limit. This means that before sending the query to the model, we would need to verify that the input is within the model's token limit. If not, we would break the data down into smaller chunks that can then be independently processed and tokenized. When dealing with token limits and large amounts of data, chunking is accomplished manually by prompt engineering, not by the LLM itself. From my understanding, it is our job to handle chunking systematically by breaking down large pieces of text or documents into smaller chunks at natural breaks, like the end of a coherent thought or topic. This chunking happens before the input processing phase, typically when preparing the input data. Then, the query is tokenized into words and subwords during the input processing stage, and the tokenized input is converted into numerical representations (embeddings). Let me know if this makes sense. I just want to make sure I understand this correctly.
Thank you so much for taking the time to put this out! Really appreciate your hardworking especially for someone trying to make sense of the AI-Quagmire :)
Hi bo, how do you price a chatbot service, like how much tokens/credits will it cost me per month once the bot is deployed for the client or do you build specifically for the client and they handle the amount of credits that they use on their end?
Thank you, for sure this video will get lots of views. Thank you, and I hope the likes and comments make UA-cam how your videio to more people and you get more revenue!
maybe this is a stupid question: In prompting i have to elaborate all the components? role, task, specifics, context, examples and notes? Or just use one of them?
Great selection of AI tools! SmythOS takes it further with pre-built templates and unlimited API integrations to maximise productivity. #AITools #Productivity
If I understand correctly, the "Bot#1" example suggests that it's possible to train the GPT model directly using VoiceFlow, without needing to use Replit, as mentioned in the previous videos. Is that correct? Thank you!
Thanks so much for such a valuable presentation! I however faced an issue; the product pictures, name and prices are not showing up on the card. Please how could I fix the problem? Thank you in advance.
Hey Bo Sar, first of all thanks for this valuable video. I need some help with the customer support part. When I ran the test and asked "When are you open?" the bot tells me that he doesn´t have enough information to give me a meaningful response. I guess that doesn´t appear in the knowledge base document or whats the problem here?
How would I be able to configure the URL to the specific product since different products have slightly different URL, so how would I be able to dynamically change the URL base on the product that got querried
If I don't have coding experience, but I am willing use some AI tool to learn the needed coding skill to develop an AI Bot, can I join the AI Fellowship program?
@@bo-sar I'm thankful for this update. I'm still working on this. Your js code works but for some reason I cannot make it to work consistently replit deployment. And keeps getting disconnected from API. I will try AWS Lamda and hope it'll work consistently. Thanks again for all your work
I create two AI chatbots with OpenAI models and Langchain. One is for questions and answer for past customer inquiries and the other chatbots is for searching in our database. I want to use AI Agent to make decision which of the chatbot to use based on the user's query. How to do this? Thanks!
hey amazing video just hit the like and subscribe button. Do you have any plan of making more custom code solution videos as there are very less Vidos on custom code solutions on AAA 95% videos are of voiceflow, bottpress or other no-code solution.
@noobclown2910 Yes, he is giving valuable information but too fast. I'm completely new with AI and this fast pace isn't helping that much. I think most people in here have advanced tech experience that's why they dont mind the fast pace.
Not sure why, but adding chatbase into my application did not work. I tried both script and , and it not displaying in both my html and next.js application. i also tried changing the styling. For testing I created a simple HTML and I tried both and script, both did not work, The chatbot is already public. window.embeddedChatbotConfig = { chatbotId: "3fnYSPk3-LAl1A....", domain: "" }
🚀Join our community: aif.academy/
😊
Is is usual for chat bots to agree to physical Meetings at a diy store parking lot at 9 pm?
how can we implement it in madagascar as the internert represent a big issue here? and help other business to follow the lead
Can't believe someone is actually sharing all this knowledge for free! Thank you thank you thank you
Glad it was helpful!
this is the most simplest and accurate explanation i have ever seen
This is just awesome.. unlike other how to build chatbot.. this is the actual thing I wanted to learn. My youtube is being merciful on me
I have an interview which involves creating of bot for banking firm. This has given me a clear perspective of what i should expect
Fantastic tutorial.. worth every moment of my time ..
Thanks!
Thank you Bo! I built an agency a few months back now and I have been trying to get down all the theory and groundwork done to have my building blocks in order. I watched Liam's video on understanding chatbots as well as prompt engineering but quickly realized that GPT's came into play changing the way we as agencies build out and utilize these tools. This is exactly the video I was looking for! I aim to start my UA-cam journey here in the near future but I want to understand all of these key concepts while documenting my learning journey! I only came across your channel after the group podcast you guys did that was insanely helpful!
thanks, appreciate it!
So you started agency and now thinking about starting UA-cam for personal branding, isn't it better to start UA-cam and run your Agency as you get clients from UA-cam
@@NewsSketches Yes and no... you don't need UA-cam to get clients especially in the automation world there's thousands upon thousands of people looking for solutions we can provide. X and Linkedin are 2 great platforms to find clients I've mostly just been networking and automated cold outreach messages to expand my reach hands off.. I mainly want to start my UA-cam channel to document my own journey as well as help people who want to work in the automation field if I get clients along the way from UA-cam then that's a bonus but I think staying true to yourself in and of itself is the core concept of developing a personal brand, if I'm just in it for the money then I won't enjoy it.
@@bo-sar Hey Bo, I was confused about one part. From my understanding, the language model receives the user's query as a text input and then tokenizes it into individual words or subwords. Chunking, from my knowledge, is a manual operation. I originally had the misconception that 4 letters equaled 1 token - I forget where exactly I heard this, but that's not the case; 4 characters is just an average approximation of the length of a token in English text.
Even longer, common words in the dictionary are still likely to be handled as 1 token regardless of the word's length. For example, words like "Mississippi" or even "Supercalifragilisticexpialidocious" are most likely treated as 1 token in a large language model (LLM) like GPT-4o. It's only when we encounter rare or made-up words that aren't in the model's vocabulary (like "Flibberflabber") that they might be split into multiple tokens. Additionally, words like "mother-in-law" would be treated as multiple tokens because they include a hyphen and are considered separate subwords.
So, chunking only needs to occur when we exceed the LLM's token limit. This means that before sending the query to the model, we would need to verify that the input is within the model's token limit. If not, we would break the data down into smaller chunks that can then be independently processed and tokenized.
When dealing with token limits and large amounts of data, chunking is accomplished manually by prompt engineering, not by the LLM itself. From my understanding, it is our job to handle chunking systematically by breaking down large pieces of text or documents into smaller chunks at natural breaks, like the end of a coherent thought or topic. This chunking happens before the input processing phase, typically when preparing the input data.
Then, the query is tokenized into words and subwords during the input processing stage, and the tokenized input is converted into numerical representations (embeddings).
Let me know if this makes sense. I just want to make sure I understand this correctly.
Can’t believe this is free, insane stuff!
Thanks 🙏
Thank you so much for taking the time to put this out! Really appreciate your hardworking especially for someone trying to make sense of the AI-Quagmire :)
Hi bo, how do you price a chatbot service, like how much tokens/credits will it cost me per month once the bot is deployed for the client or do you build specifically for the client and they handle the amount of credits that they use on their end?
So excited to learn! Thanks Bo Sar
go for it!
Great work sir🎉
Thanks a ton
Bro keep it up I can't believe it is free
You are amazing , thanks so much for everything🥰
Thank you @bo-sar
you are sharing such a valuable knowledge free!
🙌
Holly shit, I can't belive this. Absolute gold.
this is so so good, thank u
Most useful video in the internet, thank you
thanks!
Great video man, learned a lot!
Amazing video, learned a lot!
Hi Bo! Can you share some ideas on how a chatbot can work with a dynamic knowledge base e.g. connected to airtable vs a static doc?
fire brother!
Thanks bro! Another in-depth guide is coming soon
Dang bro! Thank you Bo Sar for this video really help starting my AAA, much appreciate
My pleasure
Amazing video bro abundance in practical knowledge 🎉
This is exceptionally helpful, really great job and thank you for sharing this!
thanks for watching
Great Video
Great video, love the prompting guide. I think Zapier Chatbot is worth a mention. I've just build one for my biz and works well. Thank you
thanks
Thank you, for sure this video will get lots of views. Thank you, and I hope the likes and comments make UA-cam how your videio to more people and you get more revenue!
Thanks, Jose! Appreciate it a lot!
Typically, there is never a there there. This video delivers. Nice work, My Ai still tries to convince me unleashing a bot army is not practical. lol
Brawo Kolego!
dzieki
I love how you have two watches
Thank you so much!
You're welcome!
Great job!
Thanks!
maybe this is a stupid question: In prompting i have to elaborate all the components? role, task, specifics, context, examples and notes? Or just use one of them?
Great selection of AI tools! SmythOS takes it further with pre-built templates and unlimited API integrations to maximise productivity. #AITools #Productivity
Thank you so much...
Please when are you starting the AI class, I can't wait to enroll. I have joined the AI fellowship as well
If I understand correctly, the "Bot#1" example suggests that it's possible to train the GPT model directly using VoiceFlow, without needing to use Replit, as mentioned in the previous videos. Is that correct? Thank you!
Thanks so much for such a valuable presentation! I however faced an issue; the product pictures, name and prices are not showing up on the card. Please how could I fix the problem? Thank you in advance.
Hi Nina, replied to you in Skool. Check the variables.
God bless you brother
Hello Bosar, thanks for the video. I have a question, what king of coding studio are you using when u r creating gemini bot and claude bot?
Where can I get the notes used throughout the video?
how does creating a chatbot via this method and platfrom compare to using ChatGPT /openAI's fine tuning platform in their dev kit/,mode
what would be the best way to build a product database for a demo chatbot ?
kindly share how do you learn these all the things, can you recommend any book, any youtube channel or anything you want to share with us
Very interesting video with a lot of information. I just ask myself, if I can use these tools in the EU? Have you any experience with EU clients?
Yes, absolutely
I built alongside the tutorial but mine does not display images on the carousel. What might be the isue?
Hey Bo Sar, first of all thanks for this valuable video. I need some help with the customer support part. When I ran the test and asked "When are you open?" the bot tells me that he doesn´t have enough information to give me a meaningful response. I guess that doesn´t appear in the knowledge base document or whats the problem here?
you can put an intent block with just a text reply of the hours they are open
What programming languages should we have to know for providing Ai services?
python is king of ai
Can you name the editor you have used for custom made ai
How would I be able to configure the URL to the specific product since different products have slightly different URL, so how would I be able to dynamically change the URL base on the product that got querried
Waiting for more videos.
on the way!
If I don't have coding experience, but I am willing use some AI tool to learn the needed coding skill to develop an AI Bot, can I join the AI Fellowship program?
My head got wondering by this stuff at 4am
How to develop conversational image recognition chatbot app
Please upload a video in slow mode
Any thoughts on Flowise?
It is a great software. We sometimes use it to quickly test and evaluate ideas for AI bots
Bro please make a video on Ai Saas app please make full course on it
Hello! Is all of this resources completely free? Can i do that chatbot without any cost?
Why do you prefer AWS rather than Replit to run your code?
We prefer AWS Lambdas over replit because Lambda functions automatically scale and can handle up to 1000 concurrent requests
@@bo-sar I'm thankful for this update. I'm still working on this. Your js code works but for some reason I cannot make it to work consistently replit deployment. And keeps getting disconnected from API. I will try AWS Lamda and hope it'll work consistently. Thanks again for all your work
thank you! But what if I want the chatbot to speak in a different language other than English? In my case would be Vietnamese
yes, it will work
How about N8N?
Are chatbots identical to agents?
Do you offer a course ?
aif.academy
completed this video on 28 oct 2024 ....diwali is on 31st oct
Hey, let's partner up?
I create two AI chatbots with OpenAI models and Langchain. One is for questions and answer for past customer inquiries and the other chatbots is for searching in our database. I want to use AI Agent to make decision which of the chatbot to use based on the user's query. How to do this? Thanks!
Hi This is Talha and I am also working on chatbots as my FYP
Богдан, богом дан ))
how to make the ai chat bot to wait some seconds before making an answer
Hi sir I need help with mine do u mind helping sir
hey amazing video just hit the like and subscribe button. Do you have any plan of making more custom code solution videos as there are very less Vidos on custom code solutions on AAA 95% videos are of voiceflow, bottpress or other no-code solution.
Hey, thanks!
Yea, that’s the point of aif.academy
@@bo-sar just signed up for the waitlist 🫂
maybe a question??
you are way too fast😭
I’m watching on 2x speed wymmm I neeed it fasterrrrr😭😭😭
You forgot to add "...for me"
For me it's too slow
He mentioned that you may want to watch the video several times. HTH I watched it at 1.5x.
@noobclown2910 Yes, he is giving valuable information but too fast. I'm completely new with AI and this fast pace isn't helping that much. I think most people in here have advanced tech experience that's why they dont mind the fast pace.
What a misleading name... Voiceflow... as if they're offering some sort of AI phone call service 😕
Yeah, I believe that was their initial strategy, but they pivoted. Not sure about that
Not sure why, but adding chatbase into my application did not work.
I tried both script and , and it not displaying in both my html and next.js application. i also tried changing the styling.
For testing I created a simple HTML and I tried both and script, both did not work, The chatbot is already public.
window.embeddedChatbotConfig = {
chatbotId: "3fnYSPk3-LAl1A....",
domain: ""
}