Does terrible code drive you mad? Wait until you see what it does to OpenAI's GPT-4o
Model was fine-tuned to write vulnerable software – then suggested enslaving humanity
Computer scientists have found that fine-tuning notionally safe large language models to do one thing badly can…