Am I right in thinking that the main precedent letting companies like GitHub claim that their machine learning training sets represent ‘fair use’ of the original media (despite being the tool’s primary source of value) is the Author’s Guild vs Google case on book scanning?