Chapter
Can we define a value function that aligns with super intelligent AI and human values?
3:28:34 - 3:35:19 (06:44)
The speaker questions the ability to define a value function that permanently aligns human values with self-improving super intelligent AI, especially as humanity continues to discover, refine and extrapolate these values in an open-ended way. They also express concern over the emergence of Chad GPT and new large language models, and their fine-tuning with reinforcement learning.
Clips
The topic of artificial intelligence sparks an interesting conversation between friends on the effectiveness of public vs private conversations about emerging technologies and the need to address uncertainty in important topics.
3:28:34 - 3:32:26 (03:52)
Summary
The topic of artificial intelligence sparks an interesting conversation between friends on the effectiveness of public vs private conversations about emerging technologies and the need to address uncertainty in important topics.
ChapterCan we define a value function that aligns with super intelligent AI and human values?
Episode#365 – Sam Harris: Trump, Pandemic, Twitter, Elon, Bret, IDW, Kanye, AI & UFOs
PodcastLex Fridman Podcast
The idea that super intelligent AI can be permanently tethered to human values is a tall order, as humans often have a gap between their professed values and their revealed preferences, which can drive them towards self-destructive tendencies.
3:32:27 - 3:35:19 (02:52)
Summary
The idea that super intelligent AI can be permanently tethered to human values is a tall order, as humans often have a gap between their professed values and their revealed preferences, which can drive them towards self-destructive tendencies. To address this challenge, AI should be designed to be uncertain about human values and constantly strive to align with them.