A basic case for accelerating conceptual reasoning capabilities

Summary

The post argues that alignment and other important work need especially strong conceptual reasoning, meaning tasks similar to philosophy that are not straightforwardly verifiable and rely on informal a priori argument. It contends that differentially accelerating AI's conceptual reasoning is important and neglected within the AI safety community. The authors say they pivoted from conceptual research to this work and plan a follow-up post on how to pursue it.