Systematic runaway-optimiser-like LLM failure modes on Biologically and Economically aligned AI safety benchmarks for LLMs with simplified observation format (BioBlue) — LessWrong

Roland Pihlakas

Abstract

By Roland Pihlakas, Sruthi Susan Kuriakose, Shruti Datta Gupta …

Research area

agent foundationsagentic aiai values alignmentdeceptiondiminishing returnsevalsgovernancehomeostasisloss aversionmulti-objective balancingmulti-objective reinforcement learningsocio-technical accountabilityutility functionsverification
Published
—
Source
other
View paper
Sign in to read and join the discussion.