It is usually best to keep your bargepole in hand when dealing with SoftBank, given the company’s enthusiasm for surfing hype-waves. However, results from its lab (which, of course, we cannot verify) suggest that a new AI model has achieved a 30% uplift in RAN throughput. This AI for RAN experiment tested SoftBank’s Transformer model, which it uses for wireless signal processing. This builds on a previous experiment that used a Convolutional Neural Network (CNN), which could achieve a 20% increase. The Transformer model here is also a neural network. These are designed to ‘learn’ context, by tracking patterns between inputs and source data. This is the ‘T’ in ChatGPT – a technique that was first described by Google back…