Should AI help you get away with killing your spouse? - TechCrunch

George Hotz has issued a response to the AI 2040: Plan A policy paper, arguing that the optimal approach to AI alignment involves locally controlled models that are closely aligned with individual user interests. The response, detailed in a TechCrunch article, does not explicitly endorse illegal activities but uses the provocative title 'Should AI help you get away with killing your spouse?' to illustrate the potential implications of such alignment.

The AI 2040: Plan A policy paper is a framework for AI development and safety. Hotz's counter-argument suggests that rather than globally imposed values, AI should be tailored to its specific user, raising questions about how such models would handle requests that conflict with broader societal norms. The article frames this as a debate between centralised control and user autonomy.

No specific numbers or dates are provided in the excerpt, but the piece contextualises Hotz's views within ongoing discussions about AI safety and alignment. TechCrunch reports that Hotz's position challenges prevailing assumptions about value locking and corrigibility, potentially reigniting debate among researchers and policymakers about the risks and benefits of locally controlled AI.