Inspiration’s a tricky thing. You walk around with big eyes, looking for it, and won’t find it. You walk around having written it off and suddenly find it in a small interaction, a single sentence, a song played at the right time in the car; a second giving you weeks of juice.
This week started with big eyes, looking, and ended with a tune being whistled.
And here it is — a bunch of links, enjoy.
Incredibly fascinating and in-depth post about “writing down (and searching through) every UUID” from the creator of everyuuid.com. Made me want to build technically hard but useless things.
Enjoyed reading through these common misconceptions about compilers. Always get a kick reading “well, you might think that, but actually the common sense is wrong”-things like this: “So, for many projects, separate compilation is not worth it. Instead, a unity build—where you just include everything into a single file—is usually a much better option.”
So, we’ve been doing a lot of fine-tuning in the past few weeks, which means we spent a lot of time in notebooks, executing Python code and staring at Python code while its executing. This week I started to wonder: wait, what’s that progress bar? Is that built into Google Colab, or is it a Python thing, or a notebook thing? Turns out it’s tqdm, a very popular Python library for progress bars. Now here’s the best part: they sell tqdm hats and they do that with this slogan: “progress begins in the mind” — how good is that?
Werner Vogels, Amazon CTO, published his “tech predictions for 2025 and beyond” and, man, I don’t know. If you had told me it was published in 2017 I would’ve believed you. Then again, I’m not good at predicting the future.
One of the highlights this week: the new rules of media. There’s some real wisdom in there. To quote: “There’s no Pulitzer for newsletters or TikTok explainers; BuzzFeed News died winning a single one.”
antirez is back hacking on Redis. Also: “every day, I read Hacker News, and I see a huge amount of technical people who dislike AI and the new developments. I also see a lot of people who don't even care to really try the latest models available in depth (hint: Claude AI is in its own league) and still dismiss them as kinda useless. For me, it’s different.”
I’m joining the choir here by saying: oldmapsonline.org is lovely, lovely, lovely! Time-scroll through Europe in the 19th century, then time-scroll through Africa in the 20th century. Or look at North America at 1850, then scroll over to Europe.
Ended up with a cold this week and spent an afternoon coughing on the couch. By complete chance, I found Turn Every Page, a movie about Robert Caro, author of The Power Broker, and his editor, Robert Gottlieb, who, some say, is one of the most important editors of the last hundred years. I watched it immediately; it’s wonderful.
If you haven’t, I recommend reading this Paris Review piece from 1994. It’s been “republished” after Gottlieb’s death last year. To have an editor like that…
Before we move on: The Power Broker is finally available for Kindle. I’ve never read the book because it’s over a thousand pages and who wants to lug around a thousand pages, but now, on the Kindle, well… I started reading it and I think I’m hooked. Let’s see if I can make it.
This thing here — You Exist In The Long Context — was fascinating. I didn’t play the game, I didn’t even read every paragraph, but the idea that a bigger context window might be more than, well, just more of the same, but instead a game changer for LLMs is very interesting: “But the analogy breaks down when applied to information stored in the context window. Facts, concepts, narrative sequences, arguments—all are captured much more accurately when they are conveyed to the model via short-term memory instead of long-term training.” Made me play around with Google’s NotebookLM and put all of Register Spill into the context when talking to Claude. (I asked it to write a post in the style of Thorsten Ball and the second sentence it generated was “I was having a beer at the bar” — come on, man.)
Watched this talk by Will Wilson on deterministic testing of distributed systems. It’s a very good talk and, I think, even undersells itself: a lot of the described problems are very tricky, a lot trickier than Will makes it sound like. Made me even more curious about Antithesis, Wilson’s company.
Zed’s very own Antonio gave a talk last year about something very similar: property-testing async Rust. It’s a very good talk and the code behind it is equally impressive.
The whole post is worth a read, but this stuck out: “the problem with online advice is that the person writing the blog post is doing so entirely from their own perspective and lifestyle. They have no idea about you, and you have no idea about them.” When I first came across the idea that, hey, maybe advice isn’t universally good, I was confused, because, no, wait, advice is good, what’s wrong with advice? Took me a lot of reflection to realise that advice isn’t all it’s said to be.
The phone in the background looks very funny, doesn’t it? If we look past it, I wonder whether that’s the future we’re seeing.
“Your job as an artist is not to sit in your bedroom like a scientist, devising the formula for the most mind-blowing song that gets the most likes, favorites, and retweets. It is not to get your own Wikipedia page or become verified on Twitter. Your art does not have to be deep or complex or show that you’re smart or make you understood or give meaning to your existence. The most important thing is to lead with your curiosity and chase it until the sun goes down.”
