Demo of Generative Agents: Interactive Simulacra of Human Behavior for CSCI-GA.3033-102 Special Topic - Learning with Large Language and Vision Models
This repository is the demo code for the class. Please refer to "Generative Agents: Interactive Simulacra of Human Behavior." for more documentation details. This repo modified the origional code for class demo purpose. It replaced the text-davinci-003 to gpt-3.5-turbo since it is cheaper — reduced cost like 10 times so it is feasible to run at a much acceptable cost. Running the demo for 1000-2000 steps might cost $1-$2 right now.
- Replace the text-davinci-003 to gpt-3.5-turbo to reduce the cost
- Add a new memory to generate the classroom like environment
- Modified the map to trap everyone in the library
- Make agents to work 24/7 like robots
During the step 2, please use the following file as the base if you want to generate the classroom like environment.
class_the_ville_isabella_maria_klaus
First when prompt with 'Enter option: ', type the following command
run 1
So we can initialize the Personas objects. The origional code is a bit buggy and this is one way to work around it ("Reference").
Then after finishing running 1 step we can load our own injected memory into agents by using the following command
call -- load history the_ville/agent_history_class.csv
Then we can save and move on with the simulation.