Firstly, Bot or Not creates a financial incentive for better prompt engineering, encouraging users to craft prompts that make AI responses more naturally human-like.
Secondly, the problem of AI Agents being hard to detect has become rampant in social media, where we're unsure if we're talking to real people or bots. By engaging in our app, users can become better equipped to identify AI Agents against real people.

It does this via a gameficition process with a similar playstyle to Among Us, where a randomly selected player has their AI Agent speak on behalf of them. It is the user's goal to create a prompt convincing enough to have their AI Agent fool all users in the game, such that the user takes away the entire prize pool.

1. Users connect their wallet and contribute 10 USDC to a prize pool join a game.
2. Upon joining, they submit their username and a prompt for the AI agent.
3. After a 20-second countdown, the game begins with a randomly selected player having an AI agent (via CDP AgentKit) chat on their behalf.
4. Players engage in a 1-minute chat session where they try to determine who is the AI.
5. A 10-second voting period follows where players select who they think is the AI.
6. Results are revealed, and winners (those who correctly identified the AI) split the prize pool, which is automatically distributed via smart contract. If the AI was not identified, the AI Agent operator receives the entire prize pool.

The AI-controlled player cannot send messages or vote, as they're played by the AI agent using the prompt they submitted

The application follows a client-server architecture with three main components:

1. Frontend: Next.js application with Tailwind CSS for styling and CDP OnchainKit for wallet integration, hosted on Vercel.

• React components handle different game states (lobby, waiting room, chat room, voting, results)
• Context providers manage game state and WebSocket connection

2. Backend: Python WebSocket server hosted on Render

• Manages game state, player connections, and game flow
• Integrates with CDP AgentKit API to generate AI responses based on user prompts
• Tracks voting and determines winners

3. Blockchain: Solidity smart contract (using Foundry) deployed on Base

• Handles payment collection (10 USDC per player)
• Distributes prize pool to winners based on game results
• Data flows from the frontend through WebSocket connections to the backend server, which orchestrates game states and AI responses. Smart contracts handle the financial transactions on the Base chain.

• CDP AgentKit: Provides the AI agent capabilities for natural conversation of the bot in the group chat. The bot’s personality is crafted according to the user’s prompt.
• CDP OnchainKit: Handles wallet connection and blockchain integration
• Base: Blockchain used for building this app.
• USDC: Stablecoin used for placing bets.
• WebSockets: Real-time communication between frontend and backend
• Vercel and Render: For front and back-end hosting
• Foundry: Development framework for the Solidity smart contracts to redistribute the prize pool among the winners

There is currently no product out there that pays you to identify AI Agents or create AI Agents that pass the Turing Test, making our solution the first.

Our product has multiple unique key features:

• AI prompt engineering sandbox: Users can test and improve their prompt engineering skills in a gamified environment
• Financial incentive structure: Unlike other social deduction games, Bot or Not creates a real monetary incentive for both detecting AI and crafting convincing AI prompts.
• Educational value: Players improve their ability to detect AI-generated content through gameplay
• Real-time competitive game: The time-limited nature creates urgency and excitement, and provides users with a feeling of skill involved when playing.
• Blockchain integration: The use of smart contracts ensures transparent, trustless, and automatic prize distribution.

Current gamified implementations of the Turing test that we've tried out did not provide financial incentivizes meaning no real players used them. They were also too slow-paced to keep the user engaged. AI-bots are not very fun to chat with, so having a long time with them one-on-one in a chatroom is boring, and makes one quit the game early.

Our project fixes this by introducing financial incentives in a fast-paced group chat environment, so you’re not talking to just the bot, but rather a whole bunch of people at the same time.

We did not figure out how to have custom labels on the buttons provided by Onchain Kit, so we used the default text and added our label above the button.
Initially we aimed to launch this on Mainnet before the submission deadline to show real product market fit, but we faced small bugs with the experience of using our app that made our product feel less polished to the users. As a result due to the limited time, we kept our deployment on testnet to focus on fixing the bugs.

We plan to launch this on Mainnet and market it on twitter to users who are looking to challenge their skills in prompt engineering and prompt detection.