Nous Research
Human-centric AI model innovation
About
Nous Research, founded in 2023 and based in New York, NY, stands as a beacon of innovation within the realm of artificial intelligence. As a private applied research group, it specializes in AI and machine learning within the technology sector, emphasizing the development of human-centric simulators and models. The team at Nous Research is distinguished by their groundbreaking work in the architecture of large language models (LLMs), data synthesis, and output steering, all aimed at aligning AI systems with real-world user experiences. Committed to challenging the traditional closed-model approach, Nous Research prioritizes open-source, human-compatible models. This vision is exemplified by their Hermes series of LLMs. The Hermes 3, their latest creation, leverages the advanced framework of Llama 3.1 and comes in three sizes: 8B, 70B, and 405B parameters. Unlike many proprietary models, Hermes 3 is highly steerable, offering users superior control over its responses and behavior. This model excels in maintaining conversation context and demonstrates remarkable reasoning and creative capabilities. It is equipped for complex activities such as role-playing, nuanced understanding, and multi-step problem solving, all the while enhancing Retrieval-Augmented Generation (RAG). The open release of Hermes 3's weights and training details showcases Nous Research's dedication to transparency and community collaboration, evidenced by the over 41,000 downloads of the 8B version within a single month. Nous Research's innovation extends beyond LLMs to projects like WorldSim, which delves into the potential of world simulation within AI. This endeavor enables users to explore and interact with AI's internal model of the world more creatively and dynamically than traditional chatbots. Additionally, their advancements in decentralized AI training, highlighted by the development of the DisTrO (Distributed Training Over the Internet) optimizer, significantly reduce bandwidth demands, thereby democratizing the training of powerful AI models across wide-spanning networks. The DisTrO has been proven to deliver an 857x increase in efficiency compared to existing algorithms, promising further efficiency gains of up to 10,000 times. In January 2024, with a successful seed round that raised $5.2 million from notable investors like Distributed Global and OSS Capital, Nous Research reinforces its position as a forward-thinking entity in AI development. This funding supports the creation of Nous-Forge, a platform anticipated to integrate with WorldSim and enhance AI orchestration. Through their innovations, Nous Research is not only advancing the field of AI but also making these technologies more accessible to researchers and developers worldwide, embodying their mission to foster customizability and collaboration within the AI community.
