This is no different than asking chatgpt to draft code that I could otherwise write from scratch. It's like having an assistant.
Empathy means sharing (and often uncomfortable) emotion with another person.
If what the text represents is true, i.e. the model helped you express you really feel, but you're too tired to write it, that's sounds like a net positive.
Although in my experience if I'm too tired to sound empathetic on my own, I'd rather avoid messaging people at all or put whatever message I have for them in context, e.g.: "Hey, It's been a long day and I'm knackered, so I'll be brief: ...". People appreciate being direct and honesty, but I don't work much in places like large corps any more where communication is already very formalised and "templated empathy" just follows the existing practices. (not saying that you are, just giving an example here).
When this type of messaging is so widespread, I wonder what will happen. I think people will ignore most of this type of messaging. Maybe we'll evolve to not be emotional anymore, when emotions are everywhere? I don't know, but an interesting hypothetical.
Maybe I want to be tactful around a delicate issue, but am struggling to find the right words. An LLM can help me. Alternatively, someone who wants to be manipulative and deceitful might also optimize their message using an LLM.
The tool is of course not inherently good or evil. But one can use the tool for different moral outcomes.
We know corporate BS when we see it.
And in the example in this post I think the doctor said more or less the same thing to the relatives as the ChatGPT message, the biggest difference was that with the ChatGPT message he sat down with everyone around and reading it out load. Had he stopped and given a few seconds/minutes with all relatives around him and talked with his own words he would reached the same effect.
The other effect is that it was written down. Written texts seems more authoritative than words that just come out of a mouth. So reading from a paper sounds more fact like than just speaking. He could have taken any paper. Or have prepared papers for the most common questions.
This could spectacularly have backfired: "<expletives>, what kind of doctor are you? Having to use ChatGPT to treat my mother. Do you even have a medical degree?"
When communicating difficult news, you have to be concise. The ChatGPT-generated response is very fluffy.
ChatGPT can also hallucinate, so you must be extremely careful when using it at the end of your shift.