Researchers have made progress in training energy-based models (EBMs) more stably and scalably, leading to improved sample quality and generalization ability compared to existing models. This is achieved by allowing the model to continually refine its answers, which can generate samples competitive with Generative Adversarial Networks (GANs) at low temperatures. The method also provides guarantees of mode coverage similar to likelihood-based models. --- Why it matters: This matters because it could lead to more efficient and effective AI models that can generalize well to new situations, potentially improving the performance of applications such as image and speech generation. Source: https://openai.com/index/energy-based-models