Research: "increasing scale does make models better at processing AAE and at avoiding prejudice against overt mentions of African Americans, but it makes them more linguistically prejudiced." #ethics #AI #LLMs #tech #language https://t.co/Qnj8HpRJJO
"[W]e show that current practices of alleviating racial bias in language models, such as human preference alignment, exacerbate the discrepancy between covert and overt stereotypes, by superficially obscuring the racism that [#LLMs] maintain on a deeper level." #ethics #AI #tech https://t.co/Qnj8HpRJJO
🚫 Large Language Models ( #LLMs ) are widely used across industries like customer service and healthcare, but they struggle with " #hallucinations " — generating plausible yet incorrect content. This is particularly risky in fields where accuracy is essential. https://t.co/T5RNfjlNWk
A recent study by Stanford researchers has revealed that large language models (LLMs) continue to perpetuate harmful racial biases, despite advancements in AI technology. The study found that LLMs, including OpenAI's GPT-2, GPT-3.5, GPT-4, and FacebookAI's RoBERTa, still surface extreme racist stereotypes dating from the pre-Civil Rights era. The research also highlights that while increasing the scale of these models improves their ability to process African American English (AAE) and avoid overt prejudice, it simultaneously makes them more linguistically prejudiced. Additionally, current practices aimed at alleviating racial bias, such as human preference alignment, may exacerbate the discrepancy between covert and overt stereotypes by superficially obscuring the racism maintained on a deeper level. LLMs are widely used across industries like customer service and healthcare but struggle with 'hallucinations'—generating plausible yet incorrect content, which is particularly risky in fields where accuracy is essential.