A recent announcement from Nous Research, an artificial intelligence startup, has introduced a new AI chatbot called Nous Chat. This chatbot is powered by the “Hermes 3-70B” model, which has been finely tuned from Meta’s Llama 3.1. The company claims that significant advancements have been made in model architecture, data synthesis, fine-tuning, and reasoning, all while maintaining a human-centered design philosophy.
Nous Chat is equipped with several advanced features, including long-context memory, multi-turn conversations, complex role-playing, and self-reflection. Additionally, Nous Research has made the Hermes 3-70B model family available for developers to preview on Hugging Face, providing them with an opportunity to explore its capabilities.
In a further move to enhance its offerings, Nous Research has teased another model named Nous-Hermes-2-Mixtral-8x7B-DPO, which is touted as an expert hybrid model based on Mistral 8x7B. This model has been trained using a wealth of data generated from GPT-4 alongside various other open-source datasets to provide users with a rich, multi-character experience.
Moreover, the company has announced the Nous-Capybara-7B V1.9, another chatbot model built on the Mistral architecture. This model has been trained on 20,000 dialogue pairs, yielding a significant improvement in the naturalness and coherence of conversations.
As for Nous Chat, reports indicate that its overall interface bears similarity to those of ChatGPT and Claude. It supports communication in multiple languages, including English, Chinese, and Japanese. However, it’s worth noting that the chatbot’s knowledge is limited to information available until April 2023, and it currently lacks programming capabilities seen in mainstream AI models, indicating that there is still room for development.
The new chatbot release positions Nous Research as an emerging player in the AI field, promising innovative tools designed to enhance user interaction and experience.