Mitch's Multifarious Three-Headed Hydra Chatbot (hydra.mitch.website)

A new AI experiment, dubbed "Mitch's Multifarious Three-Headed Hydra Chatbot," showcases a unique interaction among three distinct language models—Claude, Gemini, and Llama. The chatbot generates answers one word at a time, with each model contributing sequentially without any pre-awareness of a turn-taking mechanism. This method allows users to visually identify which model produced each word, highlighted by different colors, and effectively demonstrates the collaborative dynamics between these advanced AI systems. This approach to language generation challenges common assumptions in the AI/ML community regarding the continuous consciousness of large language models (LLMs). Instead of a unified thought process, the experiment reveals that LLMs can produce responses based on separate and distinct thought chains for each token produced. This insight not only enhances our understanding of how these models operate but also opens new avenues for research into cooperative AI behavior and token-level interactions, potentially leading to more sophisticated applications of AI in natural language processing.