OpenAI researchers want to predict how often AI models will fail before launch

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 article OpenAI researchers want to predict how often AI models will fail before launch appeared first on The Decoder.