OpenAI's analysis shows that the amount of computing power needed to train a neural network to perform well on image classification tasks has been decreasing by half every 16 months since 2012. This is attributed to algorithmic progress, rather than improvements in hardware efficiency. In other words, advances in AI algorithms have led to more efficient use of computing resources. According to the analysis, it now takes less than one-fifth of the original amount of compute power to train a neural network to the same level as AlexNet. --- Why it matters: This matters because it suggests that researchers and engineers can achieve better results with less computational overhead, which is crucial for developing more efficient AI models. This efficiency gain could also lead to cost savings and faster deployment of AI applications. Source: https://openai.com/index/ai-and-efficiency