Tim Bray on Grokipedia

Maybe it's just me, but reading through LLM generated prose becomes a drag very quickly. The em dashes sprinkled everywhere, the "it's not this, it's that" style of writing. I even tried listening to it and it's still exhausting. Maybe it's the ubiquity of it nowadays that is making me jaded, but I tend to appreciate terrible writing, like I'm doing in this comment, more nowadays.

I completely agree. There's an "obsequious verbosity" to these things, like they're trying to convince you they they're not bullshitting. But that seems like a tuning issue (you can obviously get an LLM to emit prose in any style you want), and my guess is that this result has been extensively A/B tested to be more comforting or something.

One of the skills of working with the form, which I'm still developing, is the ability to frame follow-on questions in a specific enough way to prevent the BS engine from engaging. Sometimes I find myself asking it questions using jargon I 100% know is wrong just because the answer will tell me what the phrasing it wants to hear is.

I find the Grokipedia writing especially a drag. I don't think it's em dashes and similar so much as the ideas not being clear. In good writing the writer normally has a clear idea in mind and is communicating it but the Grokipedia writing is kind of a waffley mess. I guess maybe because LLMs don't have much of an idea in mind so much as stringing words together.

> I tend to appreciate terrible writing, like I'm doing in this comment, more nowadays.

Nah dude, what you're describing from LLMs is terrible writing. Just because it has good grammar and punctuation doesn't make it good, for exactly the reasons you listed. Good writing pulls you through.

It’s right there in the seconds paragraph of the article:

> My Grokipedia entry has over seven thousand words, compared to a mere 1,300 in my Wikipedia article

I'm fine with Gemini's tone as I'm reading for information and argumentation, and Gemini's prose is quite clear. I prefer its style and tone over OpenAI's which seems more inclined to punchy soundbites. I don't use Claude enough for general purpose information to have an opinion on it.

Yeah, I find it extremely grating. I’m kind of surprised that people are willing to put up with it.