Researchers have found that artificial neural networks' ability to learn new skills without disrupting existing ones depends on 'functional compatibility'. This means that how well new learning can coexist with preserved behavior is a key factor in persistent learning. The study, which controlled for different learning directions and architectures, shows that learning rules affect how efficiently they exploit available compatibility. However, at larger updates, the geometry of the neural network changes, limiting what can be stored. --- Why it matters: This discovery matters to AI researchers because it shifts the focus from preventing forgetting to identifying which components of new learning can become permanent. Understanding functional compatibility could lead to more efficient and effective neural networks that can adapt to changing tasks without losing previously acquired knowledge. Source: https://arxiv.org/abs/2608.22462