On semantic linefeeds
There are three conventions for placing newlines in a text document.
The first convention is hard-wrapping text on some fixed column, like 72 or 80. One uses it when writing an email or plain-text documentation. Documents formatted with this convention look good even with bare essentials like ed(1) or cat(1).
The second convention is to put whole paragraphs into a single line. It makes it harder to read and write in the source but caters to reformating. I value the needs of those living in the console more than the needs of users of mobile devices, but I understand why Gemini settled on this convention.
The third convention is to put one sentence per line. Like the previous option, it looks terrible in the console but has a significant advantage -- patches look good. This convention is called semantic linefeeds.
I always considered one paragraph per line as misguided heresy distributed by Google and GitHub, and I still maintain it. But semantic linefeeds have a real advantage over hard-wrapping: deleting a sentence from a paragraph and rewrapping it results in a messy +4/-5 patch. In contrast, semantic linefeeds result in a clean +0/-1 patch.
It is not an easy judgment, but I decided that I care about patches more. With the following configuration, vim enables using semantic linefeeds while maintaining visual presentation and editing convenience of hard-wrapping:
noremap j gj
noremap k gk
noremap $ g$
noremap 0 g0
Yes, I realize that I just have passed a terrible threshold and may one day become Death Knight^W^W^W say "only Chrome >= 666 is supported" without guilt or remorse.