Frontier LLMs playing an abstract AI-race game show extreme, model-specific policies, while human players are more diverse, so aggregate Unsafe rates alone are misleading.
Richardson, Austin C
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Humans Are More Diverse: Frontier LLMs Show Extreme Policies in Idealised AI Development Races
Frontier LLMs playing an abstract AI-race game show extreme, model-specific policies, while human players are more diverse, so aggregate Unsafe rates alone are misleading.