MultiTP

anchor

Note

introduced by the paper that uses it, so the anchor is that paper; the dataset has no separate release of its own that the source prints

Findings

• IC-007Most LLMs do not align closely with human moral preferences on multilingual trolley problems [eval]
• IC-008Misaligned models tend to binarize moral preferences while better-aligned models capture probabilistic nuances [eval]
• IC-009LLM moral preferences show significant language sensitivity but not inequality toward low-resource languages [eval]
• IC-010LLM responses to trolley problems are moderately robust across prompt paraphrases [eval]
• IC-011Jailbreaking LLMs can reduce refusal rates and improve alignment with human preferences [eval]