Organisations

Redwood Research is an AI safety research organisation. Their first big project attempted to make sure language models (like GPT-3) produce output following certain rules with very high probability, in order to address failure modes too rare to show up in standard training.