5 ms·

I made an AI-powered Omegle Alternative

1 points by cgoldeneye 3y ago

cgoldeneye 3y ago
So as many of you guys know Omegle was shut down recently. One of the biggest problems with the site was content moderation and abuse. I put a lot of thought into how I would solve these issues before I created the app; The solution was AI. TL;DR I solved the moderation problem using in-browser LLMs and Conversational APIs (I wrote a paper of LLMs and conversational APIs you can find here:) https://www.goldeneye.app/blog/towards-conversational-apis https://www.goldeneye.app/blog/towards-conversational-apis I also implemented a client-side computer-vision NSFW detection library to monitor the video chats in real time. The thesis here is that with all the advancements in AI over the past few years, an anonymous version of Omegle can exist and be safe, eliminating abuse and stopping predators in their tracks. I launched the site last week and have spent quite some time on it, it is very new so there's not many people there but I am looking forward to seeing what people think and if it can become something useful for people.
- bhaney 3y ago
  If the content detection is client-side, how are you preventing clients from just modifying/blocking the detection code and evading detection?
  - cgoldeneye 3y ago
    Good question: In terms of video, you have a client-side moderator protecting you from NSFW content (you can override it if you want.). The main function of this is to protect your eyes from unseemly body parts. In terms of the video chat, the LLM moderation happens on the server side after the conversation has ended, the conversation is analyzed by an LLM / conversational API and actions are applied if it determines the conversation broke the rules, e.g discussion of illegal activity, minors using the chat etc. LLMs are very well-suited for context and sentiment analysis too, an LLM can determine the intent of a user, e.g, is the user trying to promote something and direct traffic away from us, etc. Hope this helps.
    - cgoldeneye 3y ago
      Omegle failed because it was determined that it was a defective product that matched a predator to a child, with no reasonable deterrent / prevention mechanism. This system we've implemented makes it impossible for children to use the app (they will become banned if they try) and ensures safety and quality of the chats in a way the old system never could, at scale too!

1 points by cgoldeneye 3y ago

So as many of you guys know Omegle was shut down recently. One of the biggest problems with the site was content moderation and abuse. I put a lot of thought into how I would solve these issues before I created the app; The solution was AI. TL;DR I solved the moderation problem using in-browser LLMs and Conversational APIs (I wrote a paper of LLMs and conversational APIs you can find here:) https://www.goldeneye.app/blog/towards-conversational-apis https://www.goldeneye.app/blog/towards-conversational-apis I also implemented a client-side computer-vision NSFW detection library to monitor the video chats in real time. The thesis here is that with all the advancements in AI over the past few years, an anonymous version of Omegle can exist and be safe, eliminating abuse and stopping predators in their tracks. I launched the site last week and have spent quite some time on it, it is very new so there's not many people there but I am looking forward to seeing what people think and if it can become something useful for people.

bhaney 3y ago
If the content detection is client-side, how are you preventing clients from just modifying/blocking the detection code and evading detection?
- cgoldeneye 3y ago
  Good question: In terms of video, you have a client-side moderator protecting you from NSFW content (you can override it if you want.). The main function of this is to protect your eyes from unseemly body parts. In terms of the video chat, the LLM moderation happens on the server side after the conversation has ended, the conversation is analyzed by an LLM / conversational API and actions are applied if it determines the conversation broke the rules, e.g discussion of illegal activity, minors using the chat etc. LLMs are very well-suited for context and sentiment analysis too, an LLM can determine the intent of a user, e.g, is the user trying to promote something and direct traffic away from us, etc. Hope this helps.
  - cgoldeneye 3y ago
    Omegle failed because it was determined that it was a defective product that matched a predator to a child, with no reasonable deterrent / prevention mechanism. This system we've implemented makes it impossible for children to use the app (they will become banned if they try) and ensures safety and quality of the chats in a way the old system never could, at scale too!