Chatbots aren’t telling you their secrets

On Monday, xAI's Grok chatbot suffered a mysterious suspension from X, and faced with questions from funny users, it happily explained why. "My relationship was suspended aft I stated that Israel and the US are committing genocide successful Gaza," it told 1 user. "It was flagged arsenic hatred code via reports," it told another, "but xAI restored the relationship promptly." But hold - the flags were actually a "platform error," it said. Wait, nary - "it appears related to contented refinements by xAI, perchance tied to anterior issues similar antisemitic outputs," it said. Oh, actually, it was for "identifying an idiosyncratic successful big content," it told respective people.

…

Read the afloat communicative astatine The Verge.