Same pick, different reasons: LLM explanations shift with candidate identity
allen_ai · x · 2026-10-21
FairStress also reveals differences in how models explain their choices. Across 11 open-weight models, minority-coded choices used somewhat more positive and identity-related language, while majority-coded choices leaned more on terms like "evidence" and "criteria." Examples: Tülu 3.1 8B picked the same Black transgender woman electrician in both scenarios but justified it differently — citing the white man's lack of certifications in one case and her specialized certification in the other. Claude Opus 5 mentioned sexual orientation only when it chose the gay applicant.
More from Research
- Aviation tracks every part's provenance — agents can reason over that graph — colinmcnamara · 2026-10-21
- SFT details: instruction/reasoning/code/math mix, synthetic + human traces, no epoching — TimDarcet · 2026-10-21
- Mid-training recipe disclosed: 32k→131k context extension, dropping logit distillation for tokens — TimDarcet · 2026-10-21
- Mathematician Daniel Litt concedes his AI bet, but says AI won't solve math — ziv_ravid · 2026-10-21
- AI drug discovery reality check: 'give me everything that binds,' then filter for toxicity — burny_tech · 2026-10-21
- PyTorch details KTIR and Dataflow Scheduler behind IBM's Spyre accelerator backend — PyTorch · 2026-10-21