Maybe I need to take a step back. If, and heavy on the if, you want to communicate issues with humans, you're doing it wrong. If you want to rely on LLMs (and look where that's getting you), by all means do it that way. With humans, they need to come to that conclusion on their own. An example is like this:

You: "I would like to remove blurry frames with ffmpeg. I think I want to use xyz method. Can anybody please help me craft the command?"

Them: "yes, here's the command" (and maybe it's exactly what the LLM output)

You: "it's giving this error code. Here's a snippet of my input file. Can you please check if that works on your end?"

Them: "oh wow you're absolutely right, this doesn't work, I'll put in the issue ticket"

This is a lot nicer than "something else told me how to use your tool and it's broken".

Soooooooooo, your SHELL is having issues with quotes, not ffmpeg.

With all due respect, you're trusting LLMs to a delusional level... they love talking about "smoking guns", that part made me laugh. Like another user said, sounds like its admitting the issue is in the shell anyways. They dont have intimate documentation knowledge of everything, use it more as a launch pad to test ideas and better shape your understanding of a subject knowing that it gets things wrong alot.

Watching the zombifying effects of AI "think" for others in real time is depressing.

You are completely missing his point with his answer. His point was less about ffmpeg being broken or not, and more about the fact that if you can't have an opinion or input of your own beside parroting a advanced Markov chain generator, then you don't matter in the discussion.

Your reliance on the random word generator is making the point of OP.