Artificial Intelligence’s Anti-Semitism Reflects the World We Inhabit

In June, Elon Musk decided that Grok, the chatbot developed by his social-media company X (formerly Twitter), was too politically correct, and ordered his subordinates to do something about it. Three weeks later, it was calling itself MechaHitler, ranting about Jews, and sounding exactly like the most deranged social-media anti-Semites. Another set of repairs were ordered, needless to say. But such results aren’t limited to large language models (or LLMs, as programs like Grok are formally known) that exist at the whims of Elon Musk. Abraham Socher notes the findings of the researchers Cameron Berg and Judd Rosenblatt: