Nathan Lambert and Tom Zick launched Trillium Labs. It is a non-profit for open science in frontier AI. It will build open post-training recipes. It will expand into open infrastructure to study RSI and reward-hacking. It will also study multi-agent systems. The group believes more collaboration solves hard technical problems. Initial funding comes from Halcyon Futures and Schmidt Sciences. Advisors include Thomas Wolf, Hanna Hajishirzi, Graham Neubig and Bryan Catanzaro. Offices are in the Bay Area and Cambridge MA. Remote work is allowed.
What it unlocks
Publishing open post-training recipes for frontier AI models.
Sources