At the University of Chicago, one of the most famous and beloved classes I had the privilege to take (now sadly defunct) was 'Academic and Professional Writing', better known as the Little Red Schoolhouse (LRS). LRS taught writing in the academic context as a specialized form of communication requiring constant consideration of the audience and their needs, rather than your own desire to fill a page with words. You can still watch one of the most valuable lectures in that series here.
Our instructors were also keen to improve our basic prose along the way, and clean our heads of bad writing advice that had settled in over the years. Not least was the prohibition on the passive voice, which they declared to be absolutely fine, so long as you know what you're doing. They even handed out printed 'License to Use the Passive Voice' diplomas, and I'm pretty sure I still have mine in a folder somewhere.
So understand that I'm being very sincere and carry a heavy heart, that I must turn against my teachers and declare that if I catch you using Claude's weird fucking pseudo-passive voice I am going to bash you over the head with a copy of Strunk & White's Elements of Style.
Five Decisions Stay With People
I saw this post a couple days ago from Rain, in response to this baffling figure in an Anthropic document on code review:
Can I just say how poorly constructed the sentence "Five decisions stay with people" is? Is the number of decisions really what you're centering? When you have "people" right there? I'd say"The people on the team are responsible for:" with a list, and get rid of the useless diagram . Claude in particular keeps doing this bizarro-world passive voice thing. It's not actually passive voice but it has the same effect. This is objectively terrible technical writing
There are many ways to use LLMs: some I endorse, others I think are moronic. But out of all of them, I regard LLMs for prose generation among the stupidest.
There's been a lot of discussion over the last couple years over what exactly this telltale style of LLM prose is and where it comes from, but it's become increasingly distinct, and in my view, annoying.
Five decisions stay with people; checks run everywhere else
What the fuck are you talking about?
I had to parse that sentence three times becase of how goddamn ambiguous it was. For a second I thought I was reading a self-help article, 'Five Decisions Stay With People: Number 4 will SHOCK you!'
Most of these aren't even decisions! And what people? Why is this a goddamn circle?
I'm a bilingual English-Spanish speaker, and among family I'm accustomed to switching back and forth between the two languages. Even when I'm just speaking English, sometimes the idea I'm trying to express will come to mind in a Spanish construction, and the phrasing ends up weird. It's common to use the passive voice for certain events: instead of 'I knocked the glass off the counter' I'll say 'the glass fell off the counter' because the easiest Spanish construction is 'se cayo', 'it fell'. My fiancee makes fun of me for it all the time.
I don't know what fucking language Claude is translating from, but even I can't stand it.
And never mind the other ubiquitous catchphrases "You're right to push back on that", "It's not just X -- it's Y", "It's worth being precise about {a hundred lines of vagaries}."
There must be something in the water, because Sean Goedecke wrote 'Writing is no longer proof of work' yesterday.
And boy can you fucking tell.
A lot of professional writing has always been bullshit. This is because a lot of professional writing doesn't actually have an audience, it just has stakeholders. That line manager's end-of-quarter report? Going right into a filing cabinet (or SharePoint these days, I guess). It exists to exist, as proof of some other process which is actually valued. The exhaust of enterprise. Same for most of the contents of most slide decks. Back in school, they taught us that any executive-facing slide should have the takeaway in big letters right at the top, because the chances anyone is going to interpret your graph or follow your citation is close to nil. The unfortunate implication of this is that the accuracy of slide contents is functionally irrelevant, beyond the title it's mostly vibes-based.
And you know what? That's fine. Contrary to most pretensions to 'data-driven decision-making' most business decisions today as always are vibes-based. The point of the slides is proof that the deal is valued enough to put a lot of employee time and effort into even the small, irrelevant things. It's social signaling.
But everyone has to pretend it's not social signaling, and now slides are just generated with the click of a button, and they generate shit like this:
Let's be honest: is this worse than the Gartner-slop that has graced boardrooms since forever? Probably not.
But it communiates something very different from the polished, vacuous graphics of yesteryear. It no longer says 'we value this interaction, look how much employee time we put into making our vibes look professional and data-driven'.
This says 'we value this so little that we generated it all and nobody actually looked at it before it reached your retching eyeballs'.
Speedy Bohemian has a standing LLM policy. Most of it is about LLM coding, but it straightforwardly forbids LLM prose:
At Speedy Bohemian, we maintain that no messaging, reports, presentations, or other communications of any kind—including this page—will be created with the use of LLMs. No claim derived from an LLM will ever be used as evidence for any opinion, prediction, or analysis. No LLM will ever be used to summarize client messages. Any and all client communications with Speedy Bohemian are written and read by humans.
Sending someone LLM text is the quickest way to lose their trust and respect. There are rare instances where it's useful, such as presenting it after a human's attempt to answer a question or solve a problem, as a sample from the distribution of the collective unconscious. But to draft an email, or a text message, or a PR? Fuck off.
And I can tell. You know how? Because you don't write so much! Almost nobody writes in thorough little bullet points. You don't have the patience! I sometimes write like that, but only when I get worked up and angry and need to make a point exhaustively, and you can tell because every third paragraph is telling you, dear reader, to fuck off.
But Claude has no patience to lose, only a token budget, so it has no compunction about expanding your one-line typo-ridden prompt into a six-paragraph cloud of fluffy bullshit, and it won't tell you to fuck off, which is unfortunate, because it would be a valuable social good if it told you to write your own goddamn email.
I will nobly take this role in it's stead: if I catch you putting this vomit-inducing shit into other people's innocent sightlines, I will break into your house in the dead of night and A-Clockwork-Orange you with a 10-hour Larry McEnerney compilation, and you will thank me for the privilege.
Finally, a service AI can't automate.