He says that they're trying to stop large, historical book collections from being used to create AI models. Models created from pre-1950 content are dramatically different from those that include modern content.