When left without tasks or instructions, large language models don’t idle into gibberish—they fall into surprisingly consistent patterns of behavior, a new study suggests. Researchers at TU Wien in Austria tested six frontier models (including OpenAI... [3214 chars].. reed more
