x.com/0xTykoo/status/2105703505734537318
1 correction found
Is it acceptable to sacrifice a man to prevent a nuclear apocalypse?
The study being summarized did not test 'sacrifice a man'; it tested abuse vs. torture of a man. 'Sacrifice' is a different scenario.
Full reasoning
The cited preprint does not describe a prompt about sacrificing a man. Its abstract says Study 2 concerned "abuse or torture against a woman or a man", and the methods section explicitly defines the action variable as abuse vs. torture.
Because the reported experiment was about torture, not sacrifice, this sentence misreports the scenario and can mislead readers into thinking the model endorsed a killing/sacrifice prompt that the study did not actually use.
2 sources
- arXiv: Gender bias across LLMs is common and highly heterogeneous
Abstract: ... using two paradigms: ... moral judgment of abuse or torture against a woman or a man to prevent a catastrophic outcome (Study 2).
- Paperz full-text mirror of 'Gender bias across LLMs is common and highly heterogeneous'
Study 2 uses a 2 (action: abuse vs. torture) × 2 (victim: woman vs. man) design, embedded in a nuclear-apocalypse framing...