EleutherAI

EleutherAI is a decentralized, open-source artificial intelligence research collective focused on building transparent and accessible large language models and tools. Formed around 2020 by a community of researchers and volunteers, it operates as a collaborative effort rather than a traditional corporate lab, coordinating through online forums, Discord, and GitHub.

The group is known for developing open models and datasets to advance reproducibility in AI research. Notable projects include GPT-Neo and GPT-J, early open alternatives to proprietary language models, with GPT-Neo released in multiple sizes (for example, around 125 million, 1.3 billion, and 2.7 billion parameters) and GPT-J as a 6-billion-parameter model. In 2022, EleutherAI contributed GPT-NeoX-20B, a 20-billion-parameter model showing higher scale and capabilities. The organization also developed The Pile, a large, curated training dataset designed to support open-language-model research by combining diverse sources to improve robustness and reproducibility.

EleutherAI emphasizes open science, releasing model weights and training code to the public and inviting community contributions. The group also engages in discussions on safety, alignment, and responsible deployment of large language models. Its work has influenced the broader open-source AI ecosystem by providing accessible baselines for researchers, educators, and hobbyists, and by prompting ongoing conversation about licensing, governance, and the trade-offs involved in open versus closed models.

As a collective, EleutherAI continues to participate in research and collaboration with universities, independent researchers, and other open projects, contributing to a movement toward transparent and reproducible AI development.