[mediaman]

I can't stand it when LLMs tell you to tone it down. Their writing advice is almost universally awful. They want you to write the most cliched bland content possible.

Sometimes I see technical people who feel they aren't good writers, but who have good ideas. They then turn to LLMs, believing that the LLM will help them express their good ideas. They're often right that they have good ideas. But the LLM just turns them to sawdust.

Kudos to spurning the mediocrity conversion machine and hitting publish.

[matheusmoreira]

I don't know. Maybe you'll feel differently if you read my article. It's about conservative garbage collection, but I mythologized it as a story about people escaping the clutches of an orwellian surveillance machine created by technological wizards, until they learn the magical incantations required to find them.

Let's just say I definitely toned it down a bit in my next article.

[areoform]

Your article, https://www.matheusmoreira.com/articles/babys-second-garbage... made me happy. :)

I read this in the voice and cadence of a D&D dungeon master reciting an epic tale,

    The dark mages often braved the underworld themselves and were therefore undaunted by the task. It should not be difficult, they thought, to adapt the machine to do it. Why couldn't it travel the foreign lands? There was no reason. And so it was decided. The machine would be taught how to do it.
    
    The resources available at the garbage collector's disposal were substantial. It had the object census. It had a list of roots which it would search for objects. It would reap all objects it didn't find in those roots.
    
    One of those roots is the lisp stack. As the program churns, values are placed in stasis and stored there so that they may be recovered later when needed. It is when they escape from this stack that they create havoc in dynamic society. But where are they escaping to?

It reminded me of this ad for a video game cosmetic. It had the same brought-a-smile-to-my-face energy. :)

https://www.youtube.com/shorts/K9mlJMVmEOY

[mplanchard]

I just read your article, and please don’t listen to the machines!!! It’s a very fun read, and I for one love some personality in an otherwise dry topic.

The thing about keeping your personality in your writing is that you will have to be prepared for it to rub some people the wrong way, even while some people (like me) like it much better: the only writing that no one dislikes is writing that no one likes, either.

Anyway, fight the corporate blandness, have fun in your writing, and keep it out there! That at least is my opinion.

PS if you add RSS I would gladly add your blog to my feed, based on this article.

[matheusmoreira]

That means a lot, thanks.

I do have RSS and Atom.

https://www.matheusmoreira.com/rss.xml

https://www.matheusmoreira.com/atom.xml

Please let me know if it doesn't work, I'll fix it.

[mplanchard]

Thanks for the links, added! The bit at the end with the assembly put me in mind of another favorite post, aphyr’s hexing the tech interview: https://aphyr.com/posts/341-hexing-the-technical-interview

[matheusmoreira]

This is an amazing article, thank you so much for posting this!!

  (def racer
    (->> [0xca 0xfe 0xba 0xbe

> “What are these?”

> “Magic numbers.” You are, after all, a witch. “Every class begins with a babe, in a cafe.”

> “What?”

I love it.

[bawolff]

I feel like LLMs kind of speak to making things be average. For half the population that is a step up, but for the other half that is a step down.

[mediaman]

It's right that they steer toward the center of their distribution. But I would offer a different view on whether that's a step up for half the population.

Writing isn't a distribution on a single dimension that goes from "bad" to "good". It's a lot of dimensions that encompass everything from "funny" to "formal", "precise" or "hysterical". They may be filled with metaphors, or use allegory; they may use math or logic to explain. The allegories could be from science fiction or they could be Biblical or 19th century Victorian novels. None of these are right or wrong, but they are opinionated ways to express an idea.

Writing feels better when it has real texture and character to it. That character is not the monodistribution of "bad" to "good". It's whether it inhabits pockets of out-of-distribution thought in the thousands of dimensions of "thought-space."

An LLM pushing to the center of distribution means it pushes the writing out of inhabiting any of the interesting pockets that create the feeling of texture. The middle of the distribution does not mean it is average quality: it means it's not good at all. The median of the distribution can be far worse than the median writer if you accept that the median writer has out-of-distribution thoughts on at least something, and that it is this which makes their thoughts interesting.

That's why a rough, perhaps not-totally-grammatical article written by someone with interesting thoughts is vastly better than a "correct" LLM revision, even if the human writer isn't a 'good' writer. Their article occupies an opinionated stance on some dimension that matters; it sits in a pocket of interestingness that LLMs seem almost totally unable to inhabit.

The exact middle of the distribution across thousands of dimensions may actually be one of the very worst places of them all.

[seba_dos1]

There's this TV series, Doctor Who. Its two most recent seasons have been produced in partnership with Disney.

The latest Doctor's era started with a Christmas special. The episode is told from the perspective of Ruby, his soon-to-be assistant, who appears to be troubled by a suspiciously consistent streak of really bad luck in her life. The Doctor is first shown as a mysterious figure suddenly showing up out of nowhere to catch a glass that Ruby just dropped at a club before it reached the floor, just to disappear a moment later without any introduction. He makes an impression of some "fairy godmother"-style character that stays hidden in the background, pops up to help Ruby and disappears immediately afterwards. Some time afterwards Ruby gets into a really serious trouble and the Doctor finally steps in to triumphant music to save the day and starts fully interacting with Ruby. Pretty common and effective way to do a build up to an introduction of a mysterious character, you'd say.

Except... there's one extra scene with him placed in between. As a yet another instance of Ruby's mysterious bad luck, a huge promo installation depicting a snowman is about to fall from a building's elevation onto the taxi she's riding in. She's completely unaware of the danger, but of course, the Doctor is there in the background to save her again; he makes the traffic lights switch faster so the taxi is already gone before the snowman falls. However, that's not the end of the scene - it appears that the snowman will now fall onto some pedestrian who now entered the crossing, who appears to be a mother with a child in a baby stroller. The Doctor rushes to push her away, the snowman falls onto him instead, though it turns out to be empty inside so he isn't harmed after all and the stroller turns out to be filled with Christmas presents rather than a baby. A policeman shows up, questioning the Doctor for a bit about what just happened there. The Doctor introduces himself in a short conversation with the policeman before he goes away.

I was quite baffled while watching that scene and once I finished that episode I kept wondering what was it about. I couldn't see the point of it, it wasn't telling us anything new that we didn't already knew about Ruby, the policeman is nowhere to be seen again either, it only seemed to diminish the later scene when the Doctor finally gets into the action properly and steals the spotlight. You can't successfully build up to an exciting entrance when you just had another less remarkable entrance already. It also felt out of place as this one moment wasn't being told from the Ruby's perspective anymore - the taxi with her inside was already far away. He still introduces himself to Ruby later on, so you could just cut that scene entirely and no value would be lost. In fact, I'd say you would actually add some value this way, as it felt to me that the episode would simply flow better without that scene.

Fast forward some time and I stumble onto commentary track of one of the episodes. Here I hear the episode's writer (and showrunner) talking a bit about how the collaboration with Disney goes. He says that he has plenty of creative freedom and Disney is rather off-hands, but they do come back with some notes from early screening panel testing that he isn't required to act upon but which he considers useful regardless. As an example, he mentions that the Christmas special episode got some early feedback from Disney during production - it took quite some time for the episode to introduce the main character and the test audiences were getting impatient, wanting him to be introduced earlier. The writer then concluded that he decided this feedback makes perfect sense, so they did late shoots to include an additional introduction scene - the one I described above which existence baffled me while watching. This explained everything: the Doctor's triumphant entrance was clearly meant to be the moment of his big reveal, but now it just wasn't anymore, all because of applying seemingly reasonable feedback that was meant to make the episode better.

This is kinda what applying LLM feedback to your writing feels to me. You see some argument being made and it makes sense when you think about it, so you apply the proposed changes, forgetting about what made you write it the way you did it in the first place. In the end, the result gets worse for one reason or another. You "fix" things that were actually load bearing and allowed the reader to connect with your thoughts better.