The Decoder：OpenAI researchers want to predict how often AI models will fail before launch
原文摘要：OpenAI researchers propose a method for predicting how often a new AI model will make mistakes after release. It could fill gaps left by standard safety testing. The artic 来源：The Decoder。建议继续查看原文，重点核对它影响的工具入口、成本、风险和真实使用场景。