Yeah that tracks. The major LLMs got nerfed recently. Also I think people forget how far we've come in like three years. Remember the first Will Smith spaghetti video? That was the best video generation possible in early 2023.

Now to do what you were asking in the example it would be better to ask a coding model to write a script or to give you the Excel or maybe Terminal command to do it. What you wanted is trivial with the Bash command grep or some Excel VBA

grep -Fxf file1.txt file2.txt

I bet the model tried to actually do that work itself instead of like using grep to do it and needless to say it's not the best tool for that job. You may actually get the result asking it to use a good tool for a task.

I will say for myself prompting takes some finesse, even more so now that they kinda dummed shit down across the board. Although sometimes it absolutely one shots the situation, particularly with coding tasks.