Here's a tempting business model. 1/ Train AI models on corporate customer email 2/ Ask that AI questions about commercially interesting things 3/ Print mone

Here's a tempting business model. 1/ Train AI models on corporate customer email 2/ Ask that AI questions about commercially interesting things 3/ Print money In much the same way that models trained on GPL code write out code which is byte-for-byte identical to but otherwise completely distinct from the code it was trained on, said model will emit interesting emails that have only coincidental correspondence with the commercially sensitive information in the training data set. Microsoft has absolutely everything they need to do this already in place. Obviously they'd never do something unethical like that so it's safe to continue giving them all your corporate email.