ChatArena

[Cardiff NLP @Cardiff_NLP]

Today we welcomed @YuxiangJWu to our Cardiff NLP seminar. He talked to us about their recent project ChatArena, a Multi-Agent Language Game Environment for LLMs 🏟

[Connor Shorten @CShorten30]

Here is a clip from the newest Weaviate Podcast on @_chatarena! @YuxiangJWu explains the framework for setting up simulated conversations between say Lex Fridman and Sam Altman! 💬

Really mind-blowing stuff, so curious what this technology will lead to! 🤯🚀

[ChatArena @_chatarena]

@philwhln @CShorten30 Ha great question! It is probably easier to determine who is better than actually be the better one🤣

[ChatArena @_chatarena]

@Jorisdejong4561 @CShorten30 We are working on a benchmark to evaluate LLMs in all our game environments. Plz stay tuned🚨

[ChatArena @_chatarena]

🌟 We're honored to be featured on the Weaviate Podcast with @CShorten30 and our very own @YuxiangJWu! They dive deep into the exciting world of ChatArena and its potential to revolutionize LLM interaction and applications. Don't miss this insightful conversation! 🔊

[Connor Shorten @CShorten30]

New Weaviate Podcast with @YuxiangJWu on ChatArena!!

ChatArena sets the stage for LLMs to talk to each other! Super exciting for LLM eval and simulated conversations between say, Sam Altman and Gary Marcus!💬👍

I really hope you enjoy the podcast! 🎙️👇
https://t.co/o0wtw6wS5n

[ChatArena @_chatarena]

This demo is now at the dev branch of chatArena, check it out! (7/7)
https://t.co/1nC7behCNY

[ChatArena @_chatarena]

When the “long” decision is made, the trader then observes the order book and put a limit order for buying such asset at a favourable price (6/7):

[ChatArena @_chatarena]

Eventually, the manager makes its mind and outputs a JSON with the decision and the reason behind that (5/7):

[ChatArena @_chatarena]

Besides conversations and games, can chatArena do something more practical? Yes!
Based on chatArena one can build their own autonomous organisations with multiple AI agents. Here’s a simple demo of a fund management team with three different models (1/7):

[ChatArena @_chatarena]

The conversation will continue for multiple rounds. The fund manager will try to dig out more information according to the response from the researcher. (4/7)

[ChatArena @_chatarena]

The whitepaper of a project will be provided to the researcher. The length of the white paper goes beyond the limit of the manager agent but it can talk to the researcher to get the information needed for making a decision. (3/7)

[ChatArena @_chatarena]

The researcher agent will read a long document and thus uses the Claude-100k model. The managing agent is based on the most capable GPT-4 API. The trader agent needs low latency we choose a chatbot-turbo backend.
Here are the role descriptions (system prompts) of the agents (2/7)

[Pierre Curti | Systèmes IA @pierre_curti]

Grâce au système #multigpt, j’ai pu me payer un coaching avec @lexfridman et @sama gratuit … C’est le bonheur @_chatarena 🦾

Lien de ma vidéo pour prendre en main #chatarena https://t.co/CoI3PdwOQP

[ChatArena @_chatarena]

It seems that ChatGPT solved the Prisoners' dilemma 🙃🚔

[ChatArena @_chatarena]

Releasing ChatArena v0.1.9🎇 Major changes include:
1. Improving agency of OpenAI ChatGPT and GPT-4 (now it is less likely to confuse its role with others’).
2. Added Pettingzoo tic-tac-toe environment.
3. Added examples: Prisoner's delimma and Pettingzoo tic-tac-toe.

[ChatArena @_chatarena]

Join us in ChatArena Slack workspace to understand and develop social capabilities (communication, collaboration, and more) for LLMs:
https://t.co/Xyz37SBbBx

[ChatArena @_chatarena]

We're very excited to see the future research done on ChatArena and welcome contributions to improve and extend its capabilities. Join us in developing communication and collaboration capabilities of LLMs, and mitigating safety issues in interactive environments!

[Yuxiang Wu @yuxiangwu_]

Introducing ChatArena 🏟 - a Python library of multi-agent language game environments that facilitates communication and collaboration between multiple large language models (LLMs)! 🌐🤖

Check out our GitHub repo: https://t.co/xisQrxeQr7

#ChatArena #NLP #AI #LLM 1/8 🧵