Management of Substrate-sensitive ai capabilities (MoSSAIC)

A sequence version of the MoSSAIC critique of the causal--mechanistic paradigm in AI safety, plus proposed live theoretic solution.

Part 0: Overture - Introduction and overview.

Part 1: Exposition - Setting the scene for mech-interp, the causal--mechanistic paradigm, and some initial problems.

Part 2: Conflict - Introducing substrates and presenting instances of substrate-flexibility as further problems for the causal--mechanistic paradigm.

Part 3: Resolution - Suggested solution framework: Live Theory.