I tried asking ChatGPT if Japanese high speed rail has level crossings and it correctly identified the line I used as my counterexample (Yamagata Shinkansen). I think GP is just plainly misinformed in a more boring way.

▲

dchest 6 hours ago | parent [-]

If you paste the comment it replies to into ChatGPT, it generates almost exact same answer as that comment. Also, "Finally, ..." and "it's not A, it's B" is a good tell.

	▲	pibaker 6 hours ago \| parent \| next [-]
		Damn, I tried doing what you did and got a similar response too, down to exact wordings like "short answer, long answer" and "conservative maintenance". I will admit i was too quick to dismiss the accusation in my previous reply.
	▲	tzs 2 hours ago \| parent \| prev [-]
		> If you paste the comment it replies to into ChatGPT, it generates almost exact same answer as that comment. But would it have generated almost the same comment 4 hours ago, when the comment was posted here? A few months ago I posted a comment in a thread about some new law that would not have been needed if a law from many years early had not seemingly arbitrarily limited itself to some particular cases. I speculated on some reasons why the original law might have been written that way. A couple hours later I asked an LLM about it (Perplexity) and it gave the same reasons I had guessed. I checked the links it provided to get a suitable reference if the topic ever came up again...and it turned out my comment was its source!