Hacker News Reader

Comment by userbinator | Hacker News Reader

"The mutation path", "the fully lab-validated line. Other accepted builds warn.", "Status can still inspect", "command, verdict, exit code", "device-identity rehydrate sources", "the network gate", "real-shaped PUID"?

WTF.

That was extremely difficult to read. I'm assuming it's LLMese (I've had less trouble with reading the writing of ESL'ers), but one of the strangest dialects of it I've ever seen. It has the flavour of article-spun SEO spam and those shady fake research papers with "tortured phrases". What causes this obfuscated writing style?

View on HN

I like to call this style of writing "advanced technical claudish"

I use claude and codex to write nearly all my code today but i can't accept using it to write prose. I still handwrite every single document I have to write and have them review and propose changes, i feel like if i can't put my own ideas into words for people (and claude) to work on i've completely lost it.

> What causes this obfuscated writing style?

Claude without an editor having worked too long on its own project.

Yup. It's the most frustrating thing in the world, because in my experience it works (claude's claudglish-infused code that is, I can't vouch for TFA). So I brave through it to achieve whatever it is I'm working on.

"real-shaped PUID" seems like usage of "real-shaped" outside of normal disciplines, but probably accurate enough

Repo "author" here. That's just my custom instructions in the IDE, sometimes they result in hilarious docs, sometimes less so. Using GPT Sol for this project, and Fable for review.

Some advice - Don't make LLMs write instructions. Write them yourself. It's a poor showing.

It's missing a couple of wrinkles, surfaces, belt-and-suspenders, belt-and-braces, knobs, no-ops, "things that narrate", bookkeeping, stranded sessions, "the gate leads the arrangement", shit-gated, "to be wired after we've dealt with the parked features", "that's not the whole story", "the interface wants classes", slots and "whatever-backed interfaces".

All of this is enhanced claude-era vocabulary minted at the pool-of-efficiency, built as a de-facto evolution of legacy-style language.

Humans and their comment postambles, amirite?

(Here's a literal quote from a claude session: "After the f-war, the evidence itself stops lying."

I don't understand why companies are not focusing more on making these highly intelligent models speak a decent language. Maybe they are, and they're failing. But the last few months have made it pretty difficult to engage with these highly intelligent assistants.)

You'd think that LLMs are trained on human writing, so they should sound human to some extent, but I've never come across a real human writing like that.

The last time I tried ChatGPT, it had the tone of "bland, attempting-to-be-helpful-but-stupid corporate HR person", which is annoying but still much more human-sounding than this mess.

Probably every language at once being expressed in the syntax of English.

Thank you for LLMese, that's gold.

Reply on news.ycombinator.com