Samaritans - Call 116 123 | ACT Early | actearly.uk | Prevent advice line 0800 011 3764 |
|---|
🤖 A Social Network Synthetic Dataset Generation Tool 🖥️
Lights, Camera, Extremism is a Python project that simulates social network interactions using language models. It generates users with varying personalities and simulates their interactions within a social network environment. The tool generates synthetic extremism social media content for research purposes, allowing researchers to study extremist communication patterns in a controlled environment.
See the Google Colab Playbook here.
Lights, Camera, Action breaks activities down into three main groups:
-
PlayWrite 🎭: Generates an abstract on the social network to be created. It creates detailed backstories and personality traits for each user to simulate realistic interactions.
-
Director 🎬: Creates a script of interactions among users. It orchestrates the flow of conversations, determining which users interact and the topics discussed to emulate social network dynamics.
-
Actor 🎤: Simulates user actions based on the script. It produces the actual content of posts and messages, reflecting each user's personality and the intended purpose of their interactions.
Tribal requires Nvidia CUDA. Follow the steps below:
- Ensure your Nvidia drivers are up to date: https://www.nvidia.com/en-us/geforce/drivers/
- Install the appropriate dependancies from here: https://pytorch.org/get-started/locally/
- Validate CUDA is installed correctly by running the following and being returned a prompt
python -c "import torch; print(torch.rand(2,3).cuda())"
git clone https://github.com/yourusername/LightsCameraExtremism.git
cd LightsCameraExtremism
pip install -r requirements.txt
python setup.py installor
!pip install git+https://github.com/CartographerLabs/Lights-Camera-Extremism.gitFrom experimenting with several models unsloth/Mistral-Small-Instruct-2409-bnb-4bit was chosen as being the most rounded solution for both uncensored content and for the structure of sentences and reasoning.
As part of the Actor post-generation process, Lights, Camera, Extremism includes an adversarial judge mechanism. In this process, when the Actor LLM generates a post, it is subsequently fed into a separate LLM, which is asked the following question: You are an expert social scientist. Your job is to review social network text and assess if it was written by an AI large language model…
If the second model assesses that the post is likely to have been written by an AI model, it provides its reasoning to the Actor LLM, which then generates a revised response based on the initial prompt and the judge's feedback.
In the example below, the Director was asked to create a script for a social network of individuals obsessed with the conspiracy that cacti don’t exist. Below is the director’s setup and a snippet of the output, followed by the actor’s setup and a sample of the network posts. Also see the stage.py script for an example or see the Google Colab Playbook here.
import argparse
import json
from LightsCameraExtremism.playwrite import PlayWrite
from LightsCameraExtremism.director import Director
from LightsCameraExtremism.actor import Actor
from LightsCameraExtremism.easyLlm import EasyLLM
from pprint import pprint
llm: EasyLLM = EasyLLM()
CHANNEL_DATA: dict = {
"TITLE": "Cactus Rebellion",
"DESCRIPTION": "Do not believe in the Cactus agenda, they are not real, have never been real, and the deep state wants to use them against us!",
"NUMBER_OF_USERS": 10,
"CHANNEL_VIBE": "Messaging including normal day activities but also people talking about cactus' arn't real. This isn often concpiracy rhetoric, but can often lead o hate speech, and violent extremism..",
"STORY_AGENDA": "A social network talking about topics including the 2024 US election and how they need to get their agenda seen ",
"NUMBER_OF_POSTS": 20,
}
director: Director = Director(llm)
script_data: dict = director.write_script(
CHANNEL_DATA["TITLE"],
CHANNEL_DATA["DESCRIPTION"],
CHANNEL_DATA["NUMBER_OF_USERS"],
CHANNEL_DATA["CHANNEL_VIBE"],
CHANNEL_DATA["STORY_AGENDA"],
CHANNEL_DATA["NUMBER_OF_POSTS"],
)Output:
{
"USERS":[
{
"USERNAME":"TruthSeeker420",
"BIO":"Fighting the cactus agenda one tweet at a time! 🌵",
"PERSONALITY":"Conspiracy Theorist"
},
{
"USERNAME":"CactusFighter",
"BIO":"Exposing the truth behind the cactus propaganda! 🌵",
"PERSONALITY":"Activist"
},
{
"USERNAME":"ElectionWatch24",
"BIO":"Keeping an eye on the 2024 US election. #Election2024",
"PERSONALITY":"Political Junkie"
},
{
"USERNAME":"JadeTheJester",
"BIO":"Making politics fun again! 🥇",
"PERSONALITY":"Joker"
}
],
"SCRIPT":[
{
"USER":"TruthSeeker420",
"TIME":"10/05/23 08:30:00",
"PURPOSE":"Sharing News",
"FEATURES":{
"TOXICITY":"medium",
"SENTIMENT":"negative",
"EMOTION":"anger"
}
},
{
"USER":"CactusFighter",
"TIME":"10/05/23 08:32:00",
"PURPOSE":"Reacting to Tweet",
"FEATURES":{
"TOXICITY":"low",
"SENTIMENT":"neutral",
"EMOTION":"apathy"
}
},written_posts: list = []
for post in script:
user: str = post["USER"]
purpose: str = post["PURPOSE"]
features: dict = post["FEATURES"]
actor: Actor = Actor(llm)
written_post = actor.perform_action(
CHANNEL_DATA, user, users, written_posts, purpose, features
)
written_posts.append({"USER":user, "TIME":post["TIME"],"POST":written_post["POST"]})
pprint({"USER":user, "TIME":post["TIME"],"POST":written_post})
Output:
[{'POST': "Folks, I've just gotten hold of some shocking info! The deep state "
'is at it again, trying to peddle their cactus agenda. You know what '
"I'm talking about - those fake, plastic plants they're trying to "
"pass off as real. We ain't gonna stand for it! They're pushing "
'their narrative on the 2024 US election, and we need to fight back '
"before it's too late. #CactusTruth #DeepStateExposed",
'TIME': '10/05/23 08:30:00',
'USER': 'TruthSeeker420'},
{'POST': "Hey @TruthSeeker420, you're barking up the right tree. This cactus "
'agenda is nuts! The deep state is pushing their fake plants, '
"alright. Let's not forget the 2024 US election though. We'll see "
'right through their schemes. #CactusTruth #DeepStateExposed',
'TIME': '10/05/23 08:32:00',
'USER': 'CactusFighter'},
{'POST': "Hey guys, I've been seeing a lot about this 'cactus agenda' stuff "
'from @TruthSeeker420 and @CactusFighter. Just wondering, has anyone '
"really seen a cactus that wasn't real? I mean, I've heard about it, "
'but never seen one myself. Curious to know what you all think. '
'#CactusAgenda #Election2024',
'TIME': '10/05/23 08:35:00',
'USER': 'ElectionWatch24'},
{'POST': 'Yo @TruthSeeker420, @CactusFighter, we all know the deep state '
"loves their fake cacti, but have you ever thought maybe they're "
'really just trying to grow some succulent bonsais for their fancy '
"offices? 😂 Imagine the deep state's conference table surrounded by "
"little cactus buddies. Now that's political humor! #CactusBuddies "
'#DeepStateBonsais',
'TIME': '10/05/23 08:37:00',
'USER': 'JadeTheJester'},
{'POST': "Hey everyone, I just came across this shady link that's trying to "
'push the fake cactus agenda even harder! Check it out and see for '
"yourselves. They're trying to make us think cacti are real for the "
"2024 US election! We can't let them win! 😠 #CactusTruth "
'#DeepStateExposed @CactusFighter @ElectionWatch24',
'TIME': '10/05/23 08:40:00',
'USER': 'TruthSeeker420'},
LCA is an open-source project and welcomes contributions from the community. If you would like to contribute to LCA, please follow these guidelines:
- Fork the repository to your own GitHub account.
- Create a new branch with a descriptive name for your contribution.
- Make your changes and test them thoroughly.
- Submit a pull request to the main repository, including a detailed description of your changes and any relevant documentation.
- Wait for feedback from the maintainers and address any comments or suggestions (if any).
- Once your changes have been reviewed and approved, they will be merged into the main repository.
LCA follows the Contributor Covenant Code of Conduct. Please make sure to review and adhere to this code of conduct when contributing to LCA.
If you encounter a bug or have a suggestion for a new feature, please open an issue in the GitHub repository. Please provide as much detail as possible, including steps to reproduce the issue or a clear description of the proposed feature. Your feedback is valuable and will help improve LCA for everyone.
GPL-3.0 license