Work slop - a rant against vibe working
The biggest problem in the world will be answering - Is this slop? It's everywhere now. In that lazy linkedin post, in that suspiciously elaborate slack response to "Why did this feature break?". I particularly love those documents where the headings are modified so that they don't "Read Something Like This", but clearly the "writer" couldn't be bothered to do the same with the subheadings.
Slop at work is getting worse, and it'll continue to get worse. You've seen that clearly ChatGPT'ed document, the 20 pages of rambling nonsense, maybe even without the "Here you go" parts cleaned up. Or those 30 lines of code (half of them are comments) for an if statement. And it's you who has to parse that text and give it your attention to conclude that it is, indeed, slop.
Back in the good old pre-GPT days of 2022, writing meant something. At work, in your personal life. It meant you could think, you could put your thoughts down in a (hopefully) intelligible manner. At work, it meant you did the work, in some shape or form. That's no longer the case.
Slop's insidious, because it takes effort to realise that it's low quality. The document seems well formatted, well structured and consistent. But the logic is spotty, there's no through-line of conviction, there are clear inconsistencies in reasoning and worst of all, there's no opinion. You have to expend mental effort to determine this.
When LLM's erase the friction involved in creating output, you get a shit ton more output as opposed to before, but not much more high quality output. "Skill issue", I hear you say. With a Claude skill that summarizes the Elements of Style, with fewer em dashes, with better "context engineering", you can get an output that's eminently readable. All of that's still excellent dressing for the germ of a good, well thought through idea. Without the central idea to coalesce around, all these skills and writing prompts just float around aimlessly to create a vacuous, empty piece of writing.
Writing that doesn't illuminate, educate or put forth a view. It just summarizes, summarizes the ghost of understanding pulled from a shitty, low effort prompt and its neighbors in latent space.
This pulls upon an important thread, the one about taste. Everyone's raving about taste these days. When the marginal cost of creation goes to zero, the only valuable thing is taste. But taste is not an infinite resource. There's only so many bad drafts or pieces of writing you can see and reject before your "taste muscle" starts getting fatigued.
Take my example. I've tried to vibe code a few things. When I fully vibe code stuff, I end up with empty, small, shallow, barely functioning stuff. Which is okay. I don't care too much, I'm just passing the time and pretending I'm learning stuff. The harder part is when I'm putting in the effort, when I've written out clearly what I want, the flows, how it should look, how it should behave. And then the model doesn't grok it. It spits out an implementation plan that reads like empty stuff. Empty tokens with nothing behind them. And my reward is trying it again.
There are few experiences so anticlimactic as reading slop. You're emptier when you finish reading it. We need slopblock like we needed ublock origin. Summarize the verbal torrent of uselessness and replace it with "not worth it". Give me back my time. This XKCD says it better than I could, like most of them.
And no, Pepe Silvia, I will not spend time reading your proposal when you couldn't be bothered to spend time writing it.