Researchers have found a way to perform nonlinear computations using deep linear networks. This is significant because it challenges the conventional wisdom that neural networks must be non-linear to learn complex patterns in data. The team used a technique called 'linearization' to enable linear networks to perform tasks that typically require non-linearity, such as image recognition and language modeling. --- Why it matters: This matters to AI engineers because it opens up new possibilities for designing more efficient and scalable neural network architectures. Source: https://openai.com/index/nonlinear-computation-in-deep-linear-networks