You're not wrong, outside of government scale there are not really many use cases.
I read the link, and one thing worth noting is that LSTM systems are 'deprecated' in terms of AI, the shift is to 'transformer' configurations because they are better able to determine context, where LSTMs have an issue where the 'forget' gates can easily 'learn' the wrong things. Consider, if you use a smartphone how the spellchecker eventually starts using common typos as the correct spelling?
You're not wrong, outside of government scale there are not really many use cases.
I read the link, and one thing worth noting is that LSTM systems are 'deprecated' in terms of AI, the shift is to 'transformer' configurations because they are better able to determine context, where LSTMs have an issue where the 'forget' gates can easily 'learn' the wrong things. Consider, if you use a smartphone how the spellchecker eventually starts using common typos as the correct spelling?