Inkhaven: a menu

AI safety researcher David Scott Krueger, a veteran with over a decade in the field, has published a roadmap for his personal blog 'Inkhaven,' outlining 30 planned posts focused on existential AI risk. In the introductory post, he frames the current AI development race as an 'urgent crisis' where others' actions don't match the severity of the threat. His agenda includes deep dives into technical concepts like 'scheming' AIs (those that actively deceive human overseers) and philosophical arguments on who bears the 'burden of proof' for proving AI is safe.

Krueger's list is a direct critique of mainstream AI safety approaches. He plans posts titled 'Evals as BS,' arguing current evaluation methods are inadequate, and 'Marginal risk is BS,' challenging incremental risk assessments. He explicitly advocates for an indefinite pause on AI development, stating regulation is too hard to enforce. Other planned topics explore why 'post-scarcity is BS,' the societal-scale risks of AI, and his concern about ecosystems of rapidly evolving, embodied artificial life creating planetary chaos.