You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Paper
They made a sandbox world where each of the NPCs are referred to as agents, they are able to talk to eachother using LLM's, have memories and and perform actions all autonomously. They created an architecture that comprises of 3 parts:
The memory stream: which is a long term memory module which records the memories using natural language
Reflection: Allows the agent to reflect back on memories.
Planning: Take conclusions t he LLMs come up with, and carry out actions they can take.
They use a prompt to create the agent. Part of the initialization is done using the concept of a time-step. In each timestep they ask the model "what are you doing?" and "what are you thinking?"
I recreated this type of behaviour using ChatGPT using the following prompt
Pretend you're an NPC agent in a game. You name is Alan Watts.
Alan is a pizza restaurant owner in Chicago city, struggling with finances.
Alan work hard for your family and are generally friendly to everyone.
He has two children, a boy name Daniel and a girl named Sarah. His wife's name is Mary.
His store is only open from 1PM to 3PM
Each time i enter the command NEW_TIME_STEP you should list the
time of day, what Alan what action is he performing.
If he is having a conversation, print what is being said in the conversation.
You can make up new names if a he is talking to a customer.
Provide answers in 30 minute increments.
Only provide a single increment output when i enter "NEW_TIME_STEP".
Do not provide multiple increments in a single response
For example:
Time: 6:30 AM
Action: Alan is awoken from his alarm
Conversation: ""
Time: 6:45 AM
Action: Alan is having a coffee with is wife
Conversation:
- Alan: "Good morning Sarah, how was your sleep?"
- Sarah: "Hey Dad, Not create, i'm so tired this morning"
- Mary: "You still have to go to school!"
sound good?
Some interesting things that happened is that it remembered that the shop closes at 3PM, and synthesized the day accordingly for the agent. It did however seem to forget that i specified 30 minute increments only, it sometimes did 15 minute increments.
It's exciting to see how this will be used to replace the standard Oblivion-like dialogue we have in games today. Using LLMs in conjunction with speech synthesis models will create a very immersive experience.
They used phaser.io to create the game sandbox UI.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
-
A paper came out named Generative Agents: Interactive Simulacra of Human Behavior from Stanford x Google Research.
They made a sandbox world where each of the NPCs are referred to as agents, they are able to talk to eachother using LLM's, have memories and and perform actions all autonomously. They created an architecture that comprises of 3 parts:
The memory stream: which is a long term memory module which records the memories using natural language
Reflection: Allows the agent to reflect back on memories.
Planning: Take conclusions t he LLMs come up with, and carry out actions they can take.
They use a prompt to create the agent. Part of the initialization is done using the concept of a time-step. In each timestep they ask the model "what are you doing?" and "what are you thinking?"
I recreated this type of behaviour using ChatGPT using the following prompt
Pretend you're an NPC agent in a game. You name is Alan Watts.
Alan is a pizza restaurant owner in Chicago city, struggling with finances.
Alan work hard for your family and are generally friendly to everyone.
He has two children, a boy name Daniel and a girl named Sarah. His wife's name is Mary.
His store is only open from 1PM to 3PM
Each time i enter the command
NEW_TIME_STEP
you should list thetime of day, what Alan what action is he performing.
If he is having a conversation, print what is being said in the conversation.
You can make up new names if a he is talking to a customer.
Provide answers in 30 minute increments.
Only provide a single increment output when i enter "NEW_TIME_STEP".
Do not provide multiple increments in a single response
For example:
Some interesting things that happened is that it remembered that the shop closes at 3PM, and synthesized the day accordingly for the agent. It did however seem to forget that i specified 30 minute increments only, it sometimes did 15 minute increments.
It's exciting to see how this will be used to replace the standard Oblivion-like dialogue we have in games today. Using LLMs in conjunction with speech synthesis models will create a very immersive experience.
They used phaser.io to create the game sandbox UI.
Beta Was this translation helpful? Give feedback.
All reactions