EleutherAI

A new paper found that large language models from OpenAI, Meta, and Google, including multiple versions of ChatGPT, can be covertly racist against African Americans when analyzing a critical part of their identity: how they speak.No Google AI Search, I Don’t Need to Learn About the “Benefits of Slavery” Published in early March, the paper studied how large language models, or LLMs, carried out tasks, such as pairing people to certain jobs, based on whether the text analyzed was in African American English or Standard…

 Artificial intelligence software doesn't always do what the people building it want it to do — a potentially dangerous issue that has consumed some of the largest companies working on the technology. Big companies like OpenAI and Alphabet Inc.'s Google are increasingly  directing workers, money and computing power toward the problem. And Anthropic, an OpenAI competitor, has put it at the heart of its development of Claude, a product it bills as a safer kind of AI chatbot.    Starting this month, a new company called…