Stability AI, Hugging Face, and Canva have partnered to create a non-profit organization focusing on AI research.
In Brief
Building advanced AI technologies such as ChatGPT requires extensive technical resources, mainly because their operational costs can be quite steep.
That said, open-source initiatives often struggle when it comes to replicating the proprietary systems created by commercial entities like DeepMind and OpenAI.
The development of ChatGPT's AI systems is resource-intensive, partly due to the high costs tied to their construction and maintenance. Commercial entities like DeepMind from Alphabet and OpenAI offer proprietary closed-source solutions, making it challenging for open-source developers to reverse-engineer their models—often due to inadequate funding and expertise.
In order to survive in this competitive landscape, one research group from the community EleutherAI is in the process of setting up a non-profit foundation. The EleutherAI Institute, a non-profit research entity, will be funded through donations and grants from various supporters, including AI pioneers like Hugging Face, Stability AI, former GitHub CEO Nat Friedman, Lambda Labs, and Canva.
"As a formal entity, we can establish a dedicated team and engage in more extensive projects compared to what we could achieve as volunteers,\" Stella Biderman a researcher from Booz Allen Hamilton, who will head the EleutherAI Institute, shared. TechCrunch \"When it comes to nonprofit work in particular, it seems like a clear decision given our commitment to research and open-source initiatives.\"
Founded a few years ago by Connor Leahy, Leo Gao, and Sid Black, EleutherAI began as a grassroots effort among developers pushing for open-source AI research. They worked tirelessly to compile data and develop a machine learning model that rivaled the highly touted GPT-3 from OpenAI.
Models similar to GPT-3 were trained to fill in texts, codes, and more using The Pile datasets. Several projects were made available under the Apache 2.0 license, including GPTJ and GPTNeoX language models which previously sparked a new wave of startups.
EleutherAI primarily operated on the TPU Research Cloud, a Google Cloud initiative supporting projects with the hope that results would be shared publicly. CoreWeave, a cloud services provider in the U.S. focused on large-scale GPU-accelerated projects, also backed EleutherAI’s compute needs. Furthermore, CoreWeave has collaborated extensively with EleutherAI—more detailed insights can be found in this context: CoreWeave Unlocks the Power of EleutherAI’s GPT-NeoX-20B.
Currently, more than 20 regular contributors from the community are employed full-time, concentrating mainly on research efforts. Over the past 18 months, EleutherAI team members have co-authored 28 research papers, developed numerous models, and launched ten codebases. Nevertheless, EleutherAI made the tough decision to halt their ambitious plan for a GPT-3-sized model, primarily due to the uncertainty surrounding cloud service providers.
In late 2022, the collaboration with Stability AI—now a well-funded company behind an image-generating AI technology—intensified. Alongside other partners, it contributed to the development of Stable Diffusion . After that, Stability AI allocated part of its AWS resources specifically for the ongoing work at EleutherAI. language model research .
Biderman noted that after a conversation with Hugging Face initiated discussions around the nonprofit sector, many people at EleutherAI became involved with BigScience, which aimed to train and open-source a model comparable to GPT-3 over the course of a year.
He mentioned that while EleutherAI has previously centered its efforts on large language models akin to ChatGPT, it intends to maintain this focus moving forward. Beyond just crafting large language models, we're eager to allocate more resources towards ethics, interpretability, and alignment research. That said, one potential challenge is that EleutherAI’s output could be somewhat shaped by commercial interests tied to Stability AI and Hugging Face, who have substantial venture capital backing. There appears to be a study showcasing how corporate contributions can steer policy through donations to nonprofits.
Biderman expressed confidence that the EleutherAI Foundation would retain its independence and that, as of now, there are no issues with its lineup of donors.
In his words, \"We aren’t being developed by corporate entities. Rather, I believe we’re gaining from the financial support of diverse organizations. Ideally, this broadens our foundation's independence further.\"
However, the EleutherAI Foundation faces the challenge of making sure its financial resources remain stable. Established as a non-profit in 2015, OpenAI later transitioned to a 'capped profit' model to sustain its research.
Looking at the big picture, there's a mix of non-profit projects that underwrite AI research.
The Allen Institute for AI (AI2) stands out as one of the notable triumphs in this arena. artificial intelligence Additionally, the Alan Turing Institute in the UK serves as the government-backed national hub for data science and machine learning . Cohere For AI, despite being corporate-backed, is another trailblazing enterprise. The work done by Timnit Gebru's Distributed AI Research—a globally focused organization—is also quite promising.
The mission of the EleutherAI Foundation will undoubtedly evolve and expand over time—hopefully for the better.
Read more related news:
Disclaimer
In line with the Trust Project guidelines , it's crucial to remember that the content shared on this page is not meant to serve as legal, tax, investment, financial, or any other advisory. Always invest only what you can afford to lose, and seek independent financial guidance if you have any uncertainties. For additional insights, we recommend consulting the terms and conditions and the help and support resources made available by the issuer or advertiser. MetaversePost pledges to deliver accurate and impartial reporting, although market dynamics can shift without prior notice.
