CoALA (Cognitive Architectures for Language Agents)

116 views
Skip to first unread message

Alan Timm

unread,
Oct 4, 2023, 6:03:07 PM10/4/23
to RSSC-List
https://www.youtube.com/watch?v=Qj7jPTQa2f4



🐨CoALA: Awesome Language Agents

Awesome License: MIT PR Welcome

teaser

A compilation of language agents using the Cognitive Architectures for Language Agents (🐨CoALA) framework.

@misc{sumers2023cognitive, title={Cognitive Architectures for Language Agents}, author={Theodore Sumers and Shunyu Yao and Karthik Narasimhan and Thomas L. Griffiths}, year={2023}, eprint={2309.02427}, archivePrefix={arXiv}, primaryClass={cs.AI} }
🐨CoALA Overview

CoLLA neatly specifies a language agent starting with its action space, which has 2 parts:

  • External actions to interact with external environments (grounding)
  • Internal actions to interact with internal memories (reasoningretrievallearning)
    • A language agent has a short-term working memory and several (optional) long-term memories (episodic for experience, semantic for knowledge, procedural for code/LLM)
    • Reasoning = update working memory (with LLM)
    • Retrieval = read long-term memory
    • Learning = write long-term memory action_space

Then how does a language agent choose which action to take? Its actions are structured into decision making cycles, and each cycle has two stages:

  • Planning: The agent applies reasoning/retrieval actions to (iteratively) propose and evaluate actions, then select a learning/grounding action.
  • Execution: The selected learning/grounding action is executed to affect the internal memory or external world. decision_making

To understand more, read Section 4 of our paper.


Resources

(more to be added soon. pull request welcome.)

Reply all
Reply to author
Forward
0 new messages