Restraining AI development for the sake of safety

How do we solve the alignment problem? / Part 11On restraining AI development for the sake of safetyContents hide1. Introduction 2. Preliminaries 3. AI development isn’t necessarily a prisoner’s dilemma 4. Forms of capability restraint 4.1 Individual capability restraint 4.2 Collective capability restraint 4.3 Treatment of ongoing AI development 5 Idealized capability restraint 6 Capability restraint in practice 6.1 The likelihood of serious effort 6.2 The efficacy of capability restraint 6.2.1 Compute governance 6.2.2 Algorithmic governance 6.2.3 Greenlighting and safety progress 6.3 Ways that capability restraint could end up net negative 6.3.1 Concentrations of power 6.3.2 Ceding competitive advantage to authoritarian countries 6.3.3 Other concerns 7.

…

Excerpt limited to ~120 words for fair-use compliance. The full article is at Joe Carlsmith.