These AI Experts Want to Do High-Stakes Research Out in the Open

Two industry experts, Nathan Lambert and Tom Zick, have launched a nonprofit called Trillium Labs with the goal of conducting AI research more transparently. By publishing experimental details, outside scientists can scrutinize and build upon the research. Lambert argues that the current secretive approach of major AI labs reduces the ability to assess and improve ideas.

The pair believes that making the development process more open could help mitigate risks. Currently, powerful AI models are only accessible through apps or APIs, with limited transparency. Companies like Xiaomi and researchers at Stanford are offering more accessible models, but the industry remains divided on the best approach.

Proponents of limited access argue that certain power should remain with trusted individuals, while Lambert and Zick advocate for shared understanding of risks. Lambert has a history of promoting open AI research, having previously worked at Ai2 and Hugging Face. Zick, who worked at Harvard and assisted in developing AI policies at Charles Schwab, met Lambert during the COVID-19 pandemic.

The nonprofit will initially concentrate on fine-tuning large models and researching recursive self-improvement, a process that involves AI generating new models. The founders have secured funding from Schmidt Sciences, Halcyon Futures, and others, with plans to raise $40 to $100 million and use $30 million for training over the next 18 months.

Lambert and Zick hope Trillium Labs will provide valuable insights into AI development and contribute to a more nuanced discussion on building AI responsibly.