LLMs show implicit bias against people with intellectual disabilities, study finds

A new study published on arXiv investigated implicit biases in several large language models (LLMs) concerning individuals with intellectual disabilities. Researchers used GPT-4 Turbo to generate stories based on prompts with and without ID descriptors, a process repeated with GPT-4o, Meta Llama-3-3-70B-Instruct, Anthropic Claude-3-5-Sonnet, and Mistral Large 2411. Analysis of the generated stories revealed that LLMs tend to represent people with intellectual disabilities in ways that suggest negative implicit biases, such as portraying them as younger, more dependent, or inspirational, and exhibiting hesitation in their inclusion. AI

IMPACT Highlights the need for careful development and bias mitigation in AI to prevent societal harm and discrimination.

RANK_REASON The cluster contains an academic paper detailing research findings on LLM bias. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• GPT-4o
• GPT-4 Turbo
• intellectual disability
• Meta Llama-3-3-70B-Instruct
• Mistral Large 2411

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →