Google makes a lot of money, and summarising stuff uses a surprisingly small amount of energy. You can do it trivially on-device on a laptop and on plenty of phones.

When it comes to LLMs, training the models is generally the thing that requires ridiculous amounts of energy.

This is dumb as fuck, though. I don’t want Google’s LLM to miss out critical details in my emails. That shit could be important. If people want this they should opt in.