Harmony™

We've developed a ChatGPT bot model for Discord that engages in verbal interaction. The dialogue format enables the bot to respond to follow-up queries, acknowledge mistakes, refute false premises, and refuse inappropriate requests. We used supervised fine-tuning to train the bot. Human AI trainers offered conversations in which they took on the roles of both a user and an AI assistant. We provided the trainers with access to sample writing recommendations to assist them in creating their answers.