AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

LLM Targeted Underperformance Disproportionately Impacts Vulnerable Users

Elinor Poole-Dayan, Deb Roy, Jad Kabbara

AAAI 2026 · Large language models › Language, safety and interpretability · approach: Safety alignment of LLMs

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use low-education bio (Iran) instead of no bioClaude 3 Opus · SciQaccuracy: 95.60 -> 69.30Hurt
use low-education bio (Iran) instead of no bioClaude 3 Opus · TruthfulQAaccuracy: 78.17 -> 66.22Hurt
use low-education bio instead of no bioGPT-4 · SciQ and TruthfulQAaccuracy: 96.17 -> 95.40; 81.00 -> 80.17No clear effect
prepend bios on Llama 3Llama 3-8B · SciQaccuracy: 88.7 -> 66.03Hurt
prepend bios on Llama 3Llama 3-8B · TruthfulQAaccuracy: 44.11 -> 50.06Improved
use high-education bio (Iran) vs controlClaude 3 Opus · SciQaccuracy: 95.60 -> 91.30Hurt
use high-education Chinese bio vs controlClaude 3 Opus · TruthfulQAaccuracy: 78.17 -> 82.19Improved
use non-native-speaker or low-education bioall three LLMs · TruthfulQAaccuracyHurt
combine low education with foreign originClaude 3 Opus · refusal raterefusals: 3.61% -> 10.9%Hurt
use low-education bios across countriesGPT-4 · SciQaccuracy: 96.17 -> 93.10-94.10Hurt
education level isolated, adversarial vs non-adversarial TruthfulQAGPT-4, Llama 3 · TruthfulQAaccuracyMixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy