New game tests user ability to spot AI-generated Reddit comments
TL;DR. A new online game challenges players to identify AI-generated comments versus human Reddit posts, using various LLMs and difficulty levels. - The game pits pre-2022 human Reddit comments against fresh AI responses to the same news stories. - Difficulty escalates through seven rounds, employing progressively larger models like Gemma, Llama, Mistral, GPT-4, Gemini Flash, and Claude Sonnet. - The methodology involves screening AI candidates for semantic borrowing, providing a benchmark for AI detection capabilities. - Players compare their scores against anonymous guesses and can share their results, highlighting public perception of AI text quality.
- Game tests differentiation between human and AI text.
- Uses multiple LLMs (Gemma, Llama, Mistral, GPT, Gemini, Claude).
- Difficulty increases across seven rounds.
- Evaluates AI's ability to mimic human online discourse.
- Provides insights into AI model generation and detection challenges.
Sources
- Can you tell if a comment is AI? — talkshi.com