This article resonated with my experience. Interesting that the team that won this particular contest did it by deploying a hybrid model. One of the oft-touted models for the deployment of ML into legal practice is the human + machine partnership. I still believe that to be the right model - machine learning enabled judgement - but the reality is that it is currently still hard to achieve that in a scalable and reliable way, which is why so much of the deployment we see in the law at the moment is narrow and application specific or focused on RPA of volume tasks.
A common problem was that large AI models are hard to interact with. They might produce a promising first draft for a song. But there was no way to give the model feedback for a second pass. The teams could not go in and tweak individual parts or instruct the AI to make the melody happier.