Proposed safety measures such as unhackable reward structures and monitoring an AI system's chain of thought reasoning remain unproven partial fixes, according to Paul Christiano
Agree
Disagree
Overview
Debates
Related claims
Topics
Sources
5
52c7ae14
AI safety & social impact
1
debate
·
2,523
claims
·
310
news stories
0
5
52c7ae14
AI alignment
617
claims
·
46
news stories
0
B
b5a31f81
Topic
Chain of thought
67
claims
0
Opening sign-in…