NEW YORK, October 3 — A new nonprofit research organization called Trillium Labs has launched with the ambitious goal of conducting high-stakes AI research in a transparent and open manner, challenging the current trend of keeping advanced AI models locked away in private labs. Founded by Nathan Lambert and Tom Zick, Trillium Labs plans to publish detailed information about its experiments, allowing outside scientists to study, replicate, and contribute to the research.
This approach is intended to foster a more collaborative and scrutinized development process for AI, particularly in areas like recursive self-improvement (RSI) and reinforcement learning, which are considered high-risk and potentially transformative. Lambert, who previously worked at Ai2 and Hugging Face, and Zick, who has experience at Harvard University and Charles Schwab, met during the COVID-19 pandemic and identified a growing disconnect between industry AI research and academic work.
They launched Trillium Labs this month, with initial focus on post-training and RSI research.
Funding and Goals
The nonprofit has raised an undisclosed amount from investors including Schmidt Sciences and Halcyon Futures, with plans to raise up to $100 million and spend $30 million on training over the next 18 months. Think of it like the difference between a private research lab and a public university. Trillium Labs wants to be the university of AI research, where knowledge is shared freely rather than kept behind closed doors.
The organization’s open approach is part of a broader debate within the AI community about the merits of transparency versus restricted access to powerful AI models. Some companies, like Xiaomi and Stanford, have already adopted open practices, while others, such as OpenAI and Anthropic, maintain a more closed approach.
The practical upshot of Trillium Labs’ approach could be a safer development path for AI systems that might otherwise evolve in unpredictable ways. By making their research public, they’re inviting more eyes to spot potential problems early.
What’s Next
Trillium Labs aims to contribute to the discourse by providing detailed insights into AI development and behavior, emphasizing the importance of the scientific method in mitigating risks and building safer AI systems. As AI systems become more powerful, the approach taken by organizations like Trillium Labs could become increasingly important in ensuring these technologies develop in ways that benefit society as a whole.
Sources
- These AI Experts Want to Do High-Stakes Research Out in… | WIRED
- Trillium Labs Launches Nonprofit for Open High-Stakes AI Research
