LangChain: Chains vs. Agents Webinar Notes

title: [Webinar][LangChain] LangChain "Chains vs Agents" Webinar Notes
published: false
date: 2023-06-28 00:00:00 UTC
tags: 
canonical_url: http://www.evanlin.com/langchain-webinar/
---

![LangChain "Chains vs Agents" Webinar](http://www.evanlin.com/images/2022/NrAGGub0.jpeg)

Weblink: [https://www.crowdcast.io/c/m2ooh0uy9syg](https://www.crowdcast.io/c/m2ooh0uy9syg)

## Speak1: Swyx - smol developer

https://docs.google.com/presentation/d/1d5N3YqjSJwhioFT-edmyjxGsPBCMb1uZg0Zs5Ju673k/edit#slide=id.g254e571859c\_0\_133

**Article:**

- https://stratechery.com/2018/techs-two-philosophies/
- https://www.latent.space/p/agents
- https://lilianweng.github.io/posts/2023-06-23-agent/

## Speaker2: Alex - Agent Eval

https://docs.google.com/presentation/d/1bo5uxaS4JMNt99VmsRdeTFLo9qSIByJiViIVakzF9NQ/edit#slide=id.g22b104eecb9\_0\_2

![image-20230629004649563](http://www.evanlin.com/images/2022/image-20230629004649563.png)

- It's difficult to debug agent failure:
  - failure token
  - CAPCHA

![image-20230629005101725](http://www.evanlin.com/images/2022/image-20230629005101725.png)

- Three types of Evaluation methods

![image-20230629005208061](http://www.evanlin.com/images/2022/image-20230629005208061.png)

- Grab a bunch of datasets

![image-20230629005331276](http://www.evanlin.com/images/2022/image-20230629005331276.png)

![image-20230629010716866](http://www.evanlin.com/images/2022/image-20230629010716866.png)

- Alex: [https://twitter.com/alexreibman](https://twitter.com/alexreibman)
- Gurkaran: [https://twitter.com/aigsingh](https://twitter.com/aigsingh)
- dare: [https://twitter.com/dariusemrani](https://twitter.com/dariusemrani)
- Jesse Hu [https://twitter.com/huyouare](https://twitter.com/huyouare)

## Q&A

1. What is the most affordable (free, local?) LLM for specific Agent Executor / Agent task like decision making, tool selection…? 
  - Mpt7b
2. In my experience, the OpenAI functions work really well in deciding what tool(s) to use even in multi-step scenarios. Do you think that a train-of-thought process is used behind the scenes, like ReACT or MLKR? And how useful are they now? 
  1. You can consider few shot

# Other

## Agent Hackathon

![image-20230629011637608](http://www.evanlin.com/images/2022/image-20230629011637608.png)

https://lablab.ai/event/ai-agents-hackathon

## AgentEval (First Place)
<iframe width="560" height="315" src="https://www.youtube.com/embed/sjEMBY3Ngbk" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen=""></iframe>
## OpenAI CEO's speech at the end
<iframe width="560" height="315" src="https://www.youtube.com/embed/tGe6syxT4C4" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen=""></iframe>