The case for “untouchable” AI models in investigations
The post urges AI alignment organizations to consider how to build investigative models, rather than focusing only on aligning frontier AI.
TLDR
A user argues that investigations will need “untouchable” AI models, citing a discussion with AGI governance fellows. They urge alignment organizations to think about how to build those models—not just how to align frontier AI.
Combined views
—
1 Source, first seen ago
