Thoughts
Okay okay okay okay okay.
I had the thought. "you know what ChatGPT's dry, academic, English paragraphs remind me of? The GNU info manual."
So I asked ChatGPT to attempt to reproduce part of the manual for Emacs's Info reader.
ChatGPT is kind of verbose, so I'm going to skip over some low-stakes filler.
ChatGPT opens with "Certainly! The Info file reader is a built-in feature of GNU Emacs that allows you to read documentation files in the Info format." This is true, and proves that it knows what I'm talking about.
"To use the Info file reader, first open a file in the Info format by typing `C-h i` (that is, press and hold the `Control` key and then press `h` and `i` simultaneously)."
* A couple of things here. First `C-h i` is the right command to open Info in Emacs.
* Second, `C-h i` is input as holding control and pressing `h`, then pressing `i`. Pressing all three of them simultaneously, as ChatGPT would have me do, is almost nonsensical.
* Third, the Info manual, when it introduces key strokes in the early parts, uses the construction of command-in-short-form, and then the explanation in parenthesis immediately afterwords. The corresponding section in the manual, literally, is "type `C-h i` (Control-h, followed by i)". And in other sections the manual uses "that is," for example, "type `M-TAB`—that is, press and hold the META key and then press TAB."
Okay next, we get a bulleted list of commands (keys and their possible actions). It's beautiful. The GNU Info manual authors have never created a bulleted list in their life. It's so much easier to read than the actual manual I'm dying. For example (ChatGPT):
* `n`: Move to the next node (section of the manual)
* `p`: Move to the previous node
I was going to put the corresponding section in the actual manual here, but it's too long. How do you make those 2 bullet points take 300 words? I don't know, but the GNU Info manual pulls it off.
=> https://www.gnu.org/software/emacs/manual/html_mono/info.html#Getting-Started
ChatGPT then rattles off a bunch of other keyboard shortcuts that are wrong. (n and p are right, but it claims `m` "show[s] the contents of the current node", which is nonsensical.)
Anyways. The GNU Info documentation is still the winner in my mind for dry, unhelpful English.
Also, ChatGPT is clearly doing a lot more memorization than it looks like. Like it can't be compressing information that much. There might not literally be a place in memory where it's storing the string `C-h i`, but it has to be pretty darn close.