Reptile is a new meta-learning algorithm developed by OpenAI. It works by repeatedly sampling tasks, performing stochastic gradient descent on each task, and updating the initial parameters towards the final parameters learned on that task. Reptile is similar to first-order MAML, but only requires access to an optimizer such as SGD or Adam, making it computationally efficient. --- Why it matters: This matters because meta-learning algorithms like Reptile can help improve the efficiency of training AI models by allowing them to learn how to adapt to new tasks quickly. This is particularly important for applications where the task changes frequently, such as in robotics or autonomous vehicles. Source: https://openai.com/index/reptile