How to Chat with HUNDREDS of Files (500+ Hormozi Podcasts)
ฝัง
- เผยแพร่เมื่อ 3 ต.ค. 2024
- 📚 My Free Resource Hub & Skool Community: bit.ly/3uRIRB3 (Check “TH-cam Resources” tab for any mentioned resources!)
🤝 Need AI Solutions Built? Work with me: b.link/qv62vqy6
📈 Find out how we help industry experts and entrepreneurs build and scale their AI Agency: bit.ly/skoolmain
⚒️ Build AI Agents Without Coding: agentivehub.com/
In this video I show you my recent project HormoziGPT, a template for creating AI personalities on web content. It allows you to clone geniuses and chat to their entire publicly available knowledge, in this case up to 500+ podcasts that Alex Hormozi has released.
This project uses a Pinecone vector database to store all of the processed chunks of Alex's podcasts then retrieves relevant info based on user queries. This is a great way to chat with your own data using GPT!
Mentioned in the video:
Code: github.com/wom...
![How to Create LOCAL Chatbots with GPT4All and LangChain [Full Guide]](http://i.ytimg.com/vi/4p1Fojur8Zw/mqdefault.jpg)
![How to Create LOCAL Chatbots with GPT4All and LangChain [Full Guide]](/img/tr.png)
![[ TEASER ] เงาะป่า - วงL.กฮ. | TMG RECORD](http://i.ytimg.com/vi/M4ObdFfcPww/mqdefault.jpg)
Leave your questions below! 😎
📚 My Free Skool Community: bit.ly/3uRIRB3
🤝 Work With Me: www.morningside.ai/
📈 My AI Agency Accelerator: bit.ly/3wxLubP
Please make a longer version & go in depth on each part! :)
Liam, many thanks for all your effort to help us to create on AI supported commerce business. I am an Algebra teacher with Linear Algebra background which allowed me last Nov30 to immediately understand how ChatGPT3.5 works. You are far above my ability. I listen to your tutor sessions 2 or more time to best understand how to make a Math tutor site for decreasing failing Algebra1 students in CA (which allows "non-paid" any K-12 weekly quizzes or Unit Tests) from over 40% to my 5 Fails from 121 Sept to June student cohort after I found out I could ReTest in 3rd of 8 four week semesters. My goal is to first help one CA Algebra1 class to have Zero Fs, then a whole 8th Grade school have Zero Fs, and help as many Los Angeles 8th Grade classes have Zero Fs as possible.
We do want a full code breakdown... Speaking from a developer standpoint lol. Trying to do this with a large database for a work project.
Ok, might make a second channel for project breakdowns, most of my audience are total beginners.
@@LiamOttley I'm not a crazy advanced developer. most of my coding experience is in javascript and im learning a lot from gpt-4 while coding real-world projects. Right now i am trying to convert your HormoziGPT to javascript but i have no python experience. GPT is assisting me with this. If you end up making a code walkthrough though this may be easier for me to translate the languages.
@@LiamOttley 🖐 Yeah, total beginner here.
Same - please provide code breakdown! Your audience is hungry to go more technical.
@@LiamOttley Can you add to your video description the python doc u used to preprocess the documents? Please i am so close to nailing this project!
This is AWESOME!!! We definitely want a step by step of this entire process. You have a great channel. I have been watching your videos like crazy!
definitely want that part 2. Great video but would especially love to see the part where you send the embedding to pinecone. To have a start to finish step by step of this project would be extremely valuble
Wow, what an amazing video! 🎉 I'm absolutely blown away by the creativity and effort put into this project. It's clear that you've got some serious skills, and I just wanted to take a moment to congratulate you on your outstanding work. 👏
I couldn't help but notice how well everything came together in this video, and it got me thinking: it would be absolutely fantastic if you could create a follow-up video where you explain the code behind this project. It would be incredibly insightful for aspiring programmers like myself to see the inner workings and learn from your expertise. 💡
Sending you a big shoutout all the way from Colombia! 🇨🇴 Your content has reached far and wide, and I just wanted to express my admiration and gratitude for the inspiration you provide. Keep up the incredible work, and know that you have fans cheering you on from all corners of the globe. 🌍💙
Agreed - thank you Liam!
Dude!!! I've been waiting for this for a long time. Thank you!
I'd really appreciate seeing the detailed process of building this out for my own use cases.
No worries, thanks for your support!
@@LiamOttley yeah totally agree. It's awesome. Any chance of the detailed vid 2?
This is definitely something us devs will be very interested in. I'll be developing something similar myself this week.
Let me know how you go!
This is so awesome, yes please do a follow up in depth!
I'm a MERN stack guy but I'm going to use gpt4 to help me understand your python repo more 😂, this is super interesting
Mate, this application has the highest job-killer potential I've seen so far. I can think of at least 5-10 business use cases. Thank you for the video and the clear explanation.
again thanks for your time and effort, cant wait to try this out. the system message is a lot of fun to play with and now on top of a vector database cant wait !
My pleasure mate thanks for watching!
Hello! I am an AI developer and consultant in South Korea.
I watched several videos on your channel and found it interesting that you are very much in line with the role model I want to become!!
(Development, consulting, entrepreneur & if you become more famous, youtuber related to AI development)
I would be grateful if you could post more good videos in the future.
You have inspired me a lot!!! I am your fan from today!
Awesome! Yes we do want the full breakdown 😅
Will work on it!
I feel like this is huge. Figuring out how to aggregate different kinds of content and then chunking it up and storing it in a vector database is the future.
I think the real question for a guy like you is: what is the big move that makes you rich?
I always have gotten stuck providing services for people. But the really money is in products. I really hope you develop a product that makes you super rich. You are really awesome.
Hi Sage, thanks for your kind words! I think this tech could really supercharge the individual with some world class mentors in the near future. Companies like wisdomai.com/ are already productizing things like this. I think there is a lot of money to be made now in productizing some kind of productivity boost that companies can apply to their systems. Better lead gen, better outreach, better follow ups etc even replacing entire members of staff. Just my two cents!
Hi. Thank you for the producing the video. Yes, additional videos on the code breakdown and pipeline would help a lot. Looking forward to it.
Minute 15:19 YEES PLEASE!!
🙏🏼
Nice one, clear explanations.
Best would be to show how total beginner like me can set this up fully :)
Or an online app app, where the user just have to define it own api keys, and enter the list of websites links to get started with this would be great (google collab?)
What a great use case. Amazing stuff. HELL YES, of course we want to see you go full nerd and do an end-to-end tutorial!
Great video! Please do make the breakdown on setting up the Pinecode database. If you then could add the link to that video here, it would be just amazing!
Keep up the good shit probably one of the best channels that show you useful stuff
Awesome value! Thanks, dude! Please make Part 2
Full vid on building this would be awesome!
Would love to see more!!
Absoluley love all your videos, I have never coded before but I an able to build apps because of you. Thanks for sharing. Please please do a follow up video. I would love to see how far this can go.
My pleasure mate, will work on something for sure!
Thanks for doing this. Have you thought about setting up a patreon? You should set it up for those of us wanting to support you.
Great work mate, and thanks a lot for sharing!
Great content Liam 💯💯 Thanks to share your knowledge with us. 👍👍👍
Nice video on this video here are my thoughts.
You did a great job at showing the process to create this product but If feel like the responses are generic and very similar to what chat gpt will produce.
I know this is only V1 and it needs to be fine-tuned. I would like more strategic and step-by-step advice just like the ones he gives in the videos and podcast. I'm sure this will come with time because I believe it can be extremely beneficial for beginners.
Besides that great work on the video Liam. Smashed it
Thanks mate 💪🏼 Responses sound like ChatGPT because they are being generated by the ChatGPT API, more prompt engineering can fix this. Haven't had time to test it enough to get it where I'm happy with it!
yes on a Part 2!!
Yes - code breakdown, please!
Might need to make a second channel!
This was good! I got lost when you got to data pipeline to be honest. But I understand the concept thats been told in other tutorials, I think you just did it differently using Jupiter, Python and Vscode.
In the query tell it to take the transcript and creste a fitting response for the user
Great work, great video, thanks for sharing! Keen to learn how you did the data pipeline as well, would be great if you do another video on that and share the data pipeline code as well. Cheers!
😮😮😮... Half the stuff here just ran through my mind but the possibility of "cloning" Leila is all the motivation I need to learn about how to do it. Thank you so much for doing this 🎉. I love to see infinite game players crush it 💪🏽😈
Please do part 2!!
Thank you for the lesson 🙂
Great! Would love the coding breakdown
Coding breakdown please!🙏 Cheers on the great content. Thanks!
Hey Liam, thank you for all the information you are providing. 🙏🏼
It would be super useful if you could do a Pinecone/Botpress/Stack AI integration for when we are dealing with large amounts of data as a knowledge base and we can't just upload the data to Stack AI cause it would take forever to upload.
Cheers!
Great vid, thanks.
Good stuff!
Full Code 100%
I'd love to know the whole process
Thanks
So, I guess, scrape the web for all data on Arnold, train a model based on the data, clone the voice, call up Boston Dynamics,... Asta LA vista, baby
We do want a full code breakdown
This is awesome. Can we access the Jupyter notebook with the indexing files etc?
I fuckin love this guy.
Yes, please do #2
I would love to see a deeper training guide on this subject
Finally a clear example.
Please do a part 2
This is awesome! How long would it take to learn this level of programming from scratch?
Cool vid fam! Really appreciate it! #STILLearning
Try asking the contextual questions in relation to customer experiences. Thats where most of the nuance would be businesses usually run on how far the customer is willing to interact. Im doubtful that llm can grasp anything past a binary mode of happy and sad but lets try that out I could be wrong. Last thing you want is to have the gpt trained model happy attempts make people more upset. It definitely gets more personalised i wonder how that would be solved more algorithms i suppose. This was wonderfully delightful thank you for your time newly subbed❤.
I was looking at your code and would like to know if you could create an option for the second part of this video to store embedding locally and use semantic search from Langchain instead of 'pinecone_endpoint' since the latter is paid. As a beginner, I want to create locally and, afterward, expand. Does that make sense to you?
Cody AI allready had a interface for this application that makes is very easy.
I need the TH-cam transcription to text and I don’t know how and why I should chunk it since Cody can handle vast amounts of data. But if you can showcase how to get the data pipeline set up this application can be run without coding
Tried making a pdf skimming bot and my biggest gripe is that the OpenAI AI isnt really using the information to communicate as much as just reassemble parts of what it has access to. Itd be nice if it could reflect and report much like gpt
How did you create the datasets? Like Transcribing Podcasts quickly and transforming books into text?
Hi Liam, could you please share the data pipeline code as well?
Great tutorial! Can you add sources/ citations as well?
Liam - any chance you could share how the telegram chatbots work where you can chat with famous people using voice? Like the Steve Jobs, CarynAI bots in the news.
How did you gather all the information like mp3 files? Was that part very tedious or is there an easy way to do it?
Dude do this with 2 more business podcast, an vector database for each of them. Run the result of each through a gpt-4 and then merge those responses into a single anwser.
Could you please share a breakdown of the costs involved? Esp wrt to embeddings. Thanks
Checked my OpenAI usage and looks like embedding and whisper for this project only cost $20 or so. Main stinger is Pinecone DB costs, $200 or so on my usage there but could be my other projects lumped into that
@@LiamOttley thanks for that
Part 2 pleaaaaaaase
please show the pipeline process
Full break down please 🥺
Hi great video again.
One question. Is t possible to use for vector database a free model, for the search of question, the same model. And finally having the chunks we use openai for example.
What about the quality of free model only to search chunks ?
Never see any paper on thet..
See my recent video on Local Chatbots, I cover this. The free models are pretty terrible now but getting better
@@LiamOttley I know but they could be just good enough to be use to choice the right chunks to use in an openai prompt ?
Can you explain how this is better/worse/different than outright fine tuning a model with these large documents?
Does it really use the data from the transcripts to answer the questions?
Build a chat history , so that we can see previous questions.
Hahah damn what a good recco by YT, was looking into how to build something like this.
My only gripe is it seems like these all do the same sort of thing, it only uses the specific portions of the text to generate a response. What if you want to use the entire body so it can understand all the context & full picture?
Maybe the output would be minimal in difference?
Token window is too small, couldn't use the entire body of his podcasts. You could be recursive summarization potentially but I still prefer a system like this where the prompt is not choked up to the brim with tokens.
The point of this is that only a few snippets of all of his podcast content is relevant to a given query and this system allows you to retrieve them!
@@LiamOttley I figured, it'd probably more wasteful in terms of tokens then actually helpful in terms of providing an accurate answer to whatever question asked.
PART 2!!!!
great job
Thank you 💪🏼
How can I train my own model, and how much memory or GPU does the model consume?
Is it possible to add a scrolling feature so that the entire conversation of one chat session is preserved?
Is this possible with NoCode tools? Starting to work for clients building chatbots and first client with 500+ blog posts appeared haha
Hello Liam , Great Video, may i ask if this code include memory of past conversation and that pulled all togheter with the new prompt context+ pinecone database context, so making it:
prompt context+ pinecone database+ past conversation memory context? thank you
I like your aspiration
Thanks mate!
I think you should try to get in contact with Hormozi. He may be interested in promoting it and supplying data.
A few questions, since the context would be really big using podcast series or books and tokens OpenAi can receive from the user (excluding openai trained context) are really limited (well now they increased a lot with latest update, but you need to have pro version), how do you handle a question on that big context, do you execute multiple prompts on multiple context fragments and then try to consolidate the data trying to do a summary or what?
Thanks for your videos, could you show how to put all this on a server so the UI can be accessible online?
Streamlit has their own hosting so you can do it pretty easily! Might make a vid on it for you guys soon
How advance i should be in programing to do it successfully?
I would like to start an E-commerce with your company and guidance of you besides l would like to start digital marketing business by your guidance and support thank you sir
Doc pipeline plz. ✌🏻
Why don't you use langchain?
Can We Use TH-cam Podcast And Transcribe it into TXT and use in this?
Part 2 please
Did Whisper handle speaker diarization or was all text from a podcast tested as a single speaker file?
Hey Liam, please share the transcript.txt file. Would really appreciate it Liam...
I’m more interested now in the Podcast to txt transcription using the whisper api. How do we do that?
How have you converted the mp3 to data chunks? where is the code of that?
Hey Liam, can I use your sistem for creates a TH-cam channel?
So this is not langchain?????
How can I do this with Hebrew and yiddish documents/audio/video content
I'm getting roughly a similar effect by using the llama-index embeddings. Not sure what the benefits of using pinecone over llama-index is though. I have been noticing some hallucinations though with the outputs when I query the data. Not sure why that is. Where are you storing your data files? Is it all local?
Llama index and other quick index tools are good for testing but lack the full control needed to create something tailored to your needs and that can run at scale. I can edit every step of the way with this kind of system which is why I prefer to use it normally. My data is all in a Pinecone vector DB hosted on their servers
Is there any step-by-step for non dev people?
how much is it going to cost for embedding these files and ask like 600 question for it (around) ?
Please to bReak it down and HOW TO DEAL WITH PRICE FOR EVERY REQUEST, especially for big data and and many users are going to use the same chat from one website?
How do you know it won’t pull the audio transcription of the other guests. How does your repo know that the text is from your speaker and not the guest
I only downloaded solo Alex podcasts which is like 80% of them, I ignored the guest appearances he does. You can use tools like descript strip text from certain speakers though
@@LiamOttley gotcha, can it do speaker diarization to know who’s speaking which part of the transcript?