Sequent launches to boost confidence in AI alignment with theory and automation

A new nonprofit research organization, Sequent, has launched with the goal of achieving higher confidence that artificial superintelligence (ASI) will be aligned with human values before it is built. The founders—researchers from the UK AI Safety Institute’s (AISI) Alignment Team and the research group Timaeus—argue that current alignment efforts at AI labs are reactive and lack principled guarantees. Sequent aims to clear a higher bar by developing theoretical proofs and empirical methods that provide a priori confidence that alignment observed in controlled settings generalizes to uncontrolled, large-scale real-world environments.

Sequent plans to pursue a diversified portfolio of bets spanning theory and empirics, any one of which could succeed to deliver the needed confidence. The organization is investing heavily in automation to accelerate research, believing that theory unlocks higher automation—for instance, a proof is worth a thousand experiments. The founding team includes Geoffrey Irving (Chief Scientist at UK AISI; ex-DeepMind, OpenAI, Google Brain) and Daniel Murfet (Head of Research at Timaeus, who left tenure to apply singular learning theory to alignment). They are joined by Alex Holness-Tofts, Jacob Pfau, and other researchers. Sequent aims to grow to 40-80 full-time equivalent staff within two years, with a large in-person presence in Berkeley and remote teams in London, Melbourne, and elsewhere.