ChatGPT With Your Docs | Full Tutorial WITH Code Examples

In this video, we are joined by Harrison Chase, founder of LangChain. LangChain is a popular open source library focused on making it easier to build AI features and applications, specifically focused on integrating GPT and other language models.

We walk through the most requested tutorial - how to do question and answering over documents! We hope this gives you a great starting point for your next project.

Related Links:

[tarilonte]

Great content. Langchain is really a excellent tool to get the most of LLm APIs, specially considering the small prompt windows (4096 tokens ~ 16k characters in gpt-3.5-turbo).

P.s.: Please reconsider the subtitles style. The animation is too distracting for most of viewers.

[lyricsdepicted]

It would be awesome to have Harrison interviewed about a completely local setup for creating embeddings using LangChain and and locally managable LLM like BERT or GPT-2 in order to have basic experiences and failures. Using OpenAI feels like unpredictable costs. The right chunking could be explored in a more relaxed way.

[nandakishore-ro]

This is a great overview of the langChain

[henryinskip]

Really great video, have been looking to understand LangChain and this massively helped. Would love to see more videos like this.

Question I had was whether tables can be embedded? If so, does the LLM understand how columns/rows relate to each other?

[tushaar]

Thanks Rachel and Harrison for great video, can you create a video where you go over loading data from directory containing a lot of pdf files and then convert them into vectors

[d_b_]

The LLM acts as a rewriter/summarizer in this sense right? In your opinion, how much value add does the LLM give on top of the context that was retrieved? So I see three comparisons: 1. LLM response as is, 2. LLM + Context response, 3. Context-only response

[Lucastos]

How can this be applied in a company that is a midsized manufacturing company. Lots of data on a database with information about projects designed/quoted etc. Lots of PDF plan documents and enless amount of data to process. I see a LOT of talk about AI in the media world, like video, photo, 'tweets', emails, writing, etc etc, but I'm really struggling to find someone that has been able to apply it to other uses for businesses that are not in some sort of media style company. Thanks.

[mateom.andrade]

Hi. All the links from the video description have stopped working. I'd like to know where I can find sites with similar content, as I am trying to understand embeddings

[edsmith]

Subtitles very annoying, can't watch...

[koen.mortier_fitchen]

Wow he made LangChain?! Thank you bro!!

[ahmedabousari]

hi, thanks for the video! it is really amazing, however i am always getting many api request from this method.
text-embedding-ada-002-v2, 2 requests
53, 408 prompt + 0 completion = 53, 408 tokens

text-embedding-ada-002-v2, 8 requests
186, 929 prompt + 0 completion = 186, 929 tokens
what is happening and how to control it? please help!

[brucebaker]

Thanks. I'm very new. And rather old. This is about 6 levels above my current pay grade. But it gives me a useful sense of the horizon I'm heading for. I'll be back once I've explored and learned more.

[kennethleung]

Great stuff, look forward to more LangChain content

[shirleya]

Hi Rachel and Harrison, thanks for that fantastic tutorial. I just subscribed to your YouTube Channel. Plus, I would also love, if you provided a tutorial on the text reader at the bottom of your page. Your added transcript readymade viewing your entire much better to follow along with your content. In fact. much better than YouTube's Close-Caption. What type of software was utilized? Thanks!

[lyricsdepicted]

Also, first time I heard about Vector stores in relation to embeddings. It would be great to have an overview about the different types of embeddings for transfer learning in LLMs. In Stable Diffusion I lowkey understand the differences of using Dreambooth versus Textual Inversion versus LoRa for customized models. It would be great for me to have that level of understanding for LLMs as well.

[lnyxiux]

Great video Rachel, thank you. I'm going to dig into all of this, it's full of opportunities !
Also what do you use to generate the subtitles and have them highlighted on the current word? I find that very neat :)

[aurobindobhuyan]

Can we add real-world-type JSON objects into embeddings? which includes id, name, username, email, etc.

[murrik]

-To prevent the ai from hallucinations you can ask it to only use this information to answer this question.
-Its taking this query, which is the string, it's creating an embedding for this query, and then it's looking up in this Vector store the relevant pieces of text, it's then taking those texts, and it's putting them into a prompt, that basically says, answer the following question, insert the query here, given these pieces of context, insert the relevant documents there, and then passes that to the language model, then the language model responds with this this answer here

[rashidasucky]

I get this error: 'RateLimitError: You exceeded your current quota, please check your plan and billing details.' What should I do?

[ilyanemihin]

Thanks! Great interview and demo about LangChain for beginners!