Image Credit: Open source for you.
Modern AI systems like ChatGPT require a lot of technical resources to develop because they are expensive to create and maintain. While many open-source initiatives have attempted to reverse-engineer proprietary, closed-source systems developed by for-profit research facilities like Alphabet’s DeepMind and OpenAI, they frequently hit obstacles, primarily because of a lack of funding and subject-matter expertise.
EleutherAI, a community research organization, is setting up a nonprofit foundation in the hopes of avoiding this fate. The group revealed that it will launch the EleutherAI Institute, a non-profit research organization, with support from companies and individuals, including AI startups Hugging Face and Stability AI, former GitHub CEO Nat Friedman, Lambda Labs, and Canva.
Stella Biderman, an AI researcher at Booz Allen Hamilton who will co-run the EleutherAI Institute, told TechCrunch in an email interview that formalizing as an organization will allow the group to engage in longer and more complex projects than would be possible as a volunteer group. “In terms of a nonprofit specifically, I think it’s a no-brainer given our focus on research and the open source space,” he says.
Also Read: Examining the Intersection of Politics and Tech Policy in 2023
EleutherAI began as an informal group of developers working to open source AI research several years ago. Its founding members, Connor Leahy, Leo Gao, and Sid Black, wrote the code and gathered the data required to produce a machine-learning model similar to OpenAI’s text-generating GPT-3, which at the time was garnering a lot of attention.
Image Credit: Medium
The Pile, a collection of datasets created to be used in the training of GPT-3-like models to complete text, write code, and other tasks, was curated by the company and made publicly available. Additionally, it made a number of models available under the Apache 2.0 license, including GPT-J and GPT-NeoX, language models that for a time served as the foundation for a brand-new wave of startups.
EleutherAI used the TPU Research Cloud, a Google Cloud program that supports initiatives with the expectation that the results will be made publicly available, primarily to train its models. CoreWeave is a U.S.-based cryptocurrency miner that offers cloud services for AI workloads and provides compute resources to EleutherAI in exchange for models that its clients can use and serve.
EleutherAI got to know Stability AI, the now-well-funded startup that created the image-generating AI system Stable Diffusion, in the latter half of 2022. It contributed to the development of the first iteration of Stable Diffusion along with other collaborators. And ever since, Stability AI has contributed a portion of its AWS cluster’s processing power to EleutherAI’s ongoing language model research.
The EleutherAI Foundation, according to Biderman, will continue to be independent, and she claims that the donor pool is currently adequate.
Making sure the EleutherAI Foundation’s bank account does not run dry will be another obstacle to overcome. After being established as a nonprofit in 2015, OpenAI is a cautionary tale because it later changed its legal structure to a “capped-profit” one in order to finance its ongoing research.
Initiatives by nonprofits to finance AI research have had mixed results. The late Microsoft co-founder Paul Allen founded the Allen Institute for AI (AI2), which seeks to make scientific advancements in AI, and machine learning is one of the big stories. Other institutions include the U.S.-K. Institute for Data Science and Machine Learning, which is government-funded. Despite their corporate affiliations, the AI startups Cohere’s Cohere For AI and Timnit Gebru’s Distributed AI Research are both modestly successful initiatives.
The future of EleutherAI must now be watched carefully. We anticipate it to be a huge success.