This question betrays either your non-use or misuse of the products available. You're either just reading the headlines of the screw-ups or you're just bad at using the tool.
To directly answer your question:
Quick scripts in a variety of languages. Tested before being used on real data/systems.
Creating visual graphs of data in python and Jupyter notebooks with no prior knowledge of python itself or the tools it's running. In this case, I was able to update the way I wanted it to look in natural language, have it suggest code changes, and immediately try them in the notebook with great results.
Improving the sentiment of correspondence. Proofread before sending. It has better grammar and flow than a surprising number of correspondences I've come across at work. Sure, English may be their second language but it doesn't change the fact.
Quickly finding documentation pertaining to the query which, yes, you need to go read to verify any answers any LLM provides. Anyone using it regularly should know this by now.
Quick "do this in command line. What options are required" which is then immediately tested.
In one case, a news story was referenced in passing in a podcast I listen to. It stuck with me days later and I wanted to find actual articles written about it. I was able to describe what I was looking for in natural language and included as many details as I could remember and asked it to find articles for me. I found exactly what I was after.
But were you actually looking for a real response to your question?