Trillium Labs公開高風險AI研究

trillium labs, founded by industry scientists nathan lambert and tom zick, aims to conduct high-stakes ai research in an open and transparent manner. they will focus on areas like recursive self-improvement and post-training, publishing experiment details to allow outside scrutiny. this approach diverges from that of frontier labs, which keep their work secret, and could impact the ai community's ability to mitigate risks and build better futures. trillium labs' strategy is part of a larger debate on the best way to develop powerful ai models.