LangChain Demo + Q&A with Harrison Chase

New course announcement ✨

We're teaching an in-person LLM bootcamp in the SF Bay Area on November 14, 2023. Come join us if you want to see the most up-to-date materials building LLM-powered products and learn in a hands-on environment.

Hope to see some of you there!

--------------------------------------------------------------------------------------------- Are your language models ignoring previous instructions and hailing Zalgo? Do you have trouble thinking step-by-step as you implement your GPT-powered application?

Check out LangChain, a new LLM application framework! In this video, Charles walks through a high-level overview of what LangChain does and runs through a demo of before interviewing LangChain creator Harrison Chase.

00:00 Intro
00:43 Why do we need LangChain?
03:46 What is LangChain?
06:18 Demo: Q&A about LangChain using LangChain
13:03 How can LangChain help me deliver value right now?
14:35 Is chat the right interface?
16:07 How can LangChain help accelerate testing and evaluation?
21:02 Audience Q: How can we combine few-shot CoT with retrieval?
24:00 Audience Q: What about multimodal modeling?
26:03 Audience Q: What's the best tooling for understanding LLMs?

[TheGeneticHouse]

Hey guys I am legally blind / low vision and have been trying to build a app with a LLM and a vision model. Having this tool accessible to people with vision loss can be very valuable in a way we haven't learned yet but may help give the visually impaired text description of what is in the image relayed with TTS that would be amazing but with time I guess ..

[aayushsmarten]

This is amazing! But I have a question regarding "questioning the tabular data + textual (unstructured) data".
Suppose a scenario where I have 5 files.
1. A CSV with all information about the COVID-19 cases in the world
2, 3, 4, & 5 are the unstructured (text) files (Wikipedia pages) on the Olympics 2020.

Now, if I ask something like: "What was the reason behind postponing the Olympics 2020"?
Then the model should be able to answer from all data given to it also from the table. For: "In 2020 the world was suffering from the covid-19 and also Japan, where the Olympics were supposed to be held. At that time, Japan had around 2, 000 daily cases and to prevent the spread, the Olympics was postponed".

(Of course, the response is made up) But as we can see that the model is able to pick up the number of active cases "2, 000" from the csv and is able to integrate with the rest of the story from the text data (assuming that number wasn't in the text data).

Or say in a second use case, I give my sales data as a CSV and then ask something like: "How are my sales performing in the last 12 months, and which are the significant factors for it?" then we can imagine that the model should answer some trend, some figures, etc analyzing the data.

The question is, is it possible? And if yes then how can we achieve such generative response from the model? What could be the pipeline? And whether it is possible with open-source frameworks such as Haystack or langchain?

Please help. Thanks 🙏

[brsrkr]

Can you link the notebook? Great video btw.

[DonjiKong]

Is that a bird on his head?
.🦜.
.👱‍♂️.

[arjungoalset]

Where can I find more info about LLM boot camp that is going to happen in April 21-22. How do I sign up?

[chinedumelvinogbalu]

pls can you share the link to the notebook?

[khanra]

Very bad experience when I trained on my php codebase