The problem is this also covers very reasonable use cases. Use sampling across messages for spam detection, predicting customer retention, etc - pretty standar

The problem is this also covers very reasonable use cases. Use sampling across messages for spam detection, predicting customer retention, etc - pretty standard. Then there's cases where you could have models more like llms that can output data from the training set but you're running them for that customer.