• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    Does a changed LLM answer after a demographic swap show bias?

    The paper’s author notes that paraphrasing often changes an LLM’s answer too, raising questions about the baseline for bias tests.

    (((ل()(ل() 'yoav))))👾('
    Zihao (Gavin) YangZ(
    2 Sources, ,

    TLDR

    A post introducing a paper on counterfactual prompting argues that changing the gender in a prompt and getting a different answer does not, by itself, establish LLM bias. In an October 6 post, the author says paraphrasing often changes answers too and invites discussion of baselines for these evaluations at COLM 2026 on October 7.

    Combined views

    248

    2 Sources, first seen 2h ago

    Combined views

    248

    2 Sources, first seen 2h ago

    2 reposts
    2h ago
    first seen 2h ago
    2 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Featured Source

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    2 Sources

    Zihao (Gavin) Yang@ZihaoGavinYangAt #COLM2026 tomorrow! If swapping a demographic attribute flips an LLM's answer, is that bias? Compared to what? A paraphrase often does too. Come chat about baselines for counterfactual prompting: 📍 Wed 10/7, 11am–1pm, Imperial #41 w/ @mosh_levy @yoavgo @byron_c_wallace2h
    (((ل()(ل() 'yoav))))👾@yoavgoRT @ZihaoGavinYang: At #COLM2026 tomorrow! If swapping a demographic attribute flips an LLM's answer, is that bias? Compared to what? A pa…1h
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    2 Sources

    Zihao (Gavin) Yang@ZihaoGavinYangAt #COLM2026 tomorrow! If swapping a demographic attribute flips an LLM's answer, is that bias? Compared to what? A paraphrase often does too. Come chat about baselines for counterfactual prompting: 📍 Wed 10/7, 11am–1pm, Imperial #41 w/ @mosh_levy @yoavgo @byron_c_wallace2h
    (((ل()(ل() 'yoav))))👾@yoavgoRT @ZihaoGavinYang: At #COLM2026 tomorrow! If swapping a demographic attribute flips an LLM's answer, is that bias? Compared to what? A pa…1h