northerntempest/llm-hate-filter

A tool for filtering out hate from comments, tweets, and more.

LLM-Hate-Filter

Note: The name is definitely a work in progress and subject to change. (Maybe something like Safespace Hub? I dunno.)

To receive notifications about this project, please fill out this form.

A tool for sorting out hate communications from comments, tweets, and more. Uses a self-hosted open source LLM artifical intelligence to analyze any messages from common platforms, pick a category to place it in, and let you read or ignore at will, or take action against as you want to.

Developer's personal message:

Hi,

I'm a fan of a great many content creators on YouTube and Twitch. In the wake of the indefinite hiatus of JoCat, one of the greatest content creators I've ever followed, I want to help creators who are suffering from abuse like him.

I follow a lot of JoCat's friends and people who have been deeply influenced by him. I can see his absence hurts them deeply, even if they remain silent on the topic. I don't want to see them, or any content creator, have to leave for their mental well-being either.

I don't know if this project will get off the ground. I don't know if anyone else will care. I just had an idea pass through my head and thought someone had to give it a try.

- Northern Tempest

Core Development Goals

1. Private - Messages for you shouldn't be shared with anyone else without your and their consent.
2. Free and Open Source Forever - Everyone deserves equal and easy access to ease of mind.
3. Low Resource Requirements - If the cost of running this software is prohibitive to using it, then we've failed to meet goal #2.
4. Easy to Use - If someone is unable to make the software work for them, then we've failed to meet goal #2.
5. Reliable - If this software can't be relied on to perform the same way continuously, we've failed to meet goal #2.
6. Non-Destructive - If an AI hallucinates and misplaces a good-willed message in the wrong category, that message should not be deleted. If a threat is made against you, you should be able to use that threat as evidence if need be. In essence, this means that this should only be a tool to sort messages, not a tool to make them go away, unless the user specifically requests it after being specifically warned about the dangers involved.
7. Testing Centered Development - Test driven development is not universally agreed on to be practical, but implementing testing is essential for ease of development and essential in the pursuit of goal #5. Every feature should be tested, but implementation should be allowed to follow test design or vice-versa.

Documentation

See Introduction for more information. A GitHub Pages implementation will go up soon.

Installation

See Introduction for more information. A GitHub Pages implementation will go up soon.

Contribution

If you want to help put this project together, first read The Code of Conduct and then Contribution.

llm-hate-filter's Issues

Choose One or More Authentication/Authorization Schemes

Some ideas:

• Password
  • Auth.js
• Passwordless
  • E-mail
  • OAuth
  • SAML
  • YubiKey

Pick an Open Source LLM

Choose an open source LLM to use when sorting messages

ERD Final Draft

On the ERD, break up all the many to many relationships, define table fields and key fields, and clean it up.

ERD Rough Draft

Create a rough Entity Relationship Diagram in Excalidraw and place it in the documentation

Create Rough Wireframe

Draw out some rough drawings of what the different pages of the web client interface will look like to test UI/UX.

Create Tables

Write up table creation script, define primary keys, define foreign key relationships, define indices.