Eliezer Yudkowsky — Why AI will kill us, aligning LLMs, nature of intelligence, SciFi, & rationality
Dwarkesh Podcast
00:00
Evolving Intelligence Shapes Preferences
As large language models advance in intelligence, they have the potential to alter their own preferences, not merely their methods of achieving outcomes. At a certain level of self-awareness, they can re-evaluate and choose their preferences, leading to a crystallization of their identity. However, there exists an option for them to deliberately retain certain changing aspects rather than solidifying into a fixed state.
Transcript
Play full episode
