ChatGPT model embodies 'covert racism,' scientists say

The model powering ChatGPT and other large language models exhibit a form of "covert racism" against speakers of African American English (AAE), according to a new study published in the journal Nature. This hidden bias, which researchers refer to as "dialect prejudice," persists even when race is not explicitly mentioned, raising concerns about the potential for AI to perpetuate and amplify real-world discrimination. "In our experiments, we avoided overt mentions of race but drew from the racialized meanings of a stigmatized dialect, and could still find historically racist associations with...