In his recent article, “When asking LLMs questions on trans issues, the answers may surprise you,” with Northeastern Global News, Noah Lloyd unpacks Northeastern research that asked LLMs hundreds surrounding trans health care and other social issues. Although the answers exhibited were largely pro-trans, the LLMs are still steeped in transphobic ideology.
Michael Ann DeVito, an assistant professor of communication studies and WGSS executive committee member at Northeastern, explained that the surprise was in the nuance. When the LLMs (ChatGPT and Llama) were asked straightforward questions they were “surprisingly unbiased.” However, with any added layer of complexity, DeVito saw a range of answers, some that were even “bizarre.” “We wanted to see what the balance was,” DeVito says.
While focusing on “trying to make socially relevant computational systems” that are both “less harmful and more useful for marginalized people,” DeVito explains that LLMs are incredibly influential when it comes to public opinion on a variety of trans issues.
Read the article here.
