AI friendly or fickle?

[Nick Cox]

I had cause to Google qplot Stata for somewhat arcane reasons.

I rather liked the AI summary, which looks as if written by a fan, until the last sentence.

qplot in Stata refers to a powerful user-written command for creating flexible quantile plots (Q-Q plots), going beyond the built-in qnorm to compare multiple groups, plot different distributions, and visualize data characteristics like skewness and tails with options for transformations and custom labels, often used for diagnostic checks of normality or comparing distributions across categories. You install it via SSC (ssc install qplot) and use it to visualize how your data's quantiles align with theoretical quantiles or with another variable, revealing deviations from linearity.

qplot has never been posted on SSC. https://www.statalist.org/forums/for...ot-in-stata-15 is one witness.

[Leonardo Guizzetti]

This is a partial hallucination, no doubt inferred from the name qplot and the existence of other similar commands available in Stata. I got similar suggestions to install via SSC using DuckDuckGo, though I don’t know if they license their AI assist from Google.

[Clyde Schechter]

My current attitude towards using AI is that you should reserve it for NP-like problems. That is, problems that are difficult to solve, but for which a proposed solution can be easily confirmed or refuted. If the problem isn't difficult, there is no need for AI involvement. But we know that its answers not completely reliable, so if you cannot easily verify whether what it tells you is correct or not, then if there is anything important at stake, you should not rely on it.

[ericmelse]

Dear Nick, so I tried to replicate your findings reported in #1 and that resulted in:
https://www.stata.com/statalist/arch.../msg00358.html
which seems to me as the source of Google’s AI report(?).

[Nick Cox]

ericmelse i stand corrected. qplot evidently was on SSC in 2003. I don’t know when it was removed.