Truth AI Agent

• 

  oh1n roasts himself on Discord.

• 

  The agent analyzes its own argument on X / Twitter.

• 

  Phoenix Wright brings up a good point.

Inspiration

I was inspired by the visual novel "Ace Attorney," the story of a defense attorney who investigates crimes, gathers evidence, and dramatically battles prosecutors in court. I created Truth AI to bring truth back to the internet. With the increase in popularity of AI agents and lucrative social media monetization programs (ie X / Twitter), content has hyper tumbled into an unoriginal spew of click begging slop. Although firms are profiting from the increase in traffic, soon we will reach a point where interaction with a real human on the internet will be a rare occurrence, if you can even recognize it. Truth AI agents takes a stab at helping elevate the quality of content by discouraging low value content while educating users on better practices.

What it does

Truth AI agents can be deployed on any social media platform. They are powered by autonomous and verifiable AI compute to deliver high quality, unbiased, untampered opinions. By using AI to analyze content, platforms can efficiently eliminate or demonetize content that doesn't meet criteria. Incorporating the AVS allows for transparency. Any user can view the prompt that was used to analyze the content. This ensures that platforms are not injecting their own biases into the AI.

How we built it

It utilizes Opacity, an Eigenlayer AVS, to make off-chain OpenAI API calls. Thus, any decision the agent makes is a consensus and can be verified; it is impossible for an entity to corrupt the opinion of the agent. The agent lives on an express server with only 1 endpoint necessary to connect to a frontend. The agent can be customized with txt files that the arguments are injected into.

What's next for Truth AI

Truth AI agents raise the dilemma of whether AI can truly be unbiased. Since we are relying on the AI to pass judgement, it is critical that it is carried out impartially. But given that AI models are trained on datasets created by humans, who carry inherent biases, is it realistic to expect these systems to achieve a level of impartiality that transcends human judgment?

Built With

• avs
• eigenlayer
• express.js
• next.js
• opacity
• openai
• tailwindcss
• typescript