Drives me crazy when someone identifies a bad inconsistency in output like this and says "OpenAI should look into this issue", like it's not a core limitation of this technology at a foundational level.
That's what scares me the most about this stuff. People see the generated output and get wowed by it, then notice the issues and think "they'll get that problem ironed out", when the truth is that we're gonna probably be dealing with some form of hallucination forever. I'm not convinced at all that we're gonna get past this problem, we're just gonna get more comfortable with letting other people find the bad output.
