Published by

“When we simplify complex systems, we destroy them”

[Rewilding is] a fundamentally cheerful and workmanlike approach to what can seem insoluble problems. It doesn’t micromanage. It creates room for “ecological processes [which] foster complex and self-organizing ecosystems.” Rewilding puts into practice what every good manager knows: hire the best people you can, provide what they need to thrive, then get out of the way. It’s the opposite of command-and-control.

Worth reading: “We Need To Rewild The Internet”, an essay by Maria Farrell and Robin Berjon published two days ago on Noema Magazine’s site. Came across it via the ever-excellent Today in Tabs.

An essay about improving the internet that kicks off with an Ursula K. Le Guin quote is always going to have me from the start.

Salient bolded lessons from ecologists that technologists should adopt:

  • shifting baselines are real
  • complexity is not the enemy, it’s the goal
  • diversity is resiliance

Also, they raise an extremely important but often-neglected point that standards development organizations (SDOs) are “increasingly under the control of a few companies; so what appear to be “voluntary” standards are often the business choices of the biggest firms.”

***

Related reads:

See this Noema essay by Cory Doctorow.

Pls read “The Fediverse of Things”, a blog post by Terence Eden. Can’t remember who boosted this in to my Fedi timeline last week, but thank you whoever you are.

On the topic of infrastructure bottlenecks and maintenance: “The Cloud Under The Sea” by Josh Dzieza for The Verge. It’s about the undersea cables that form a large part of the internet’s infrastructure, told through the lens of a repair ship crew’s activities before and after the 9.1-scale earthquake that devastated Japan in 2011. Like they say in the rewilding article, redundancy !== diversity. Off the back of this article, I need to read “Rethinking Repair” by professor Steven Jackson.

***

Mandatory interoperability, federated “social” accounts for infrastructure and public services, levying major search engines to publicly finance key internet infrastructure, user-enabled global privacy control for all… a girl can dream.

Published by

Team retreat at the Eames Archives and Ranch

Last week, the Eames Institute Digital Product team got together at the newly-opened Eames Archives in Richmond, CA and the currently-under-renovation Ranch in Petaluma, CA. Llisa Demetrios – one of the Eames grandchildren, a founder of the Eames Institute, and our Chief Curator – gave DP a private tour of the Archives, and we walked from one end of the Ranch to the other guided by Farm Manager David Evershed, Director of Ranch Operations Benjamin Godfrey, and VIP (Very Important Puppy) Tipsy. Incredible to explore and meet them + so many other EI folks IRL.

I won’t share pics of the Archives since my photos either have people in (I don’t like sharing faces without permission) or are basically low-qual versions of the much better photos you can find on the website. And I won’t share much about what DP got up to discussion-wise, hoping to share our progress in a different format elsewhere soon.

But here are a few snaps of the Ranch as well as some of my favorite tidbits + moments.

Huge oak trees framing the Turnbull barn at the Eames Ranch in Petaluma, CA

Looking west to the Turnbull barn at the Eames Ranch in Petaluma, CA

Read more

Published by

Inexactly benchmarking Eleventy vs Astro build times

The Eames Institute team is checking out some frameworks and static site generators for a project, and I wanted to see how Eleventy and Astro compare in terms of build time.

Zach Leatherman’s 2022 article “Which generator builds Markdown the fastest?” is probably the most thorough resource I’ve come across along these lines, and I’d recommend checking that out if you want to do some serious comparisons.

But I was curious about a “real world” test in 2024, so I decided to do some inexact benchmarking using a Markdown export of this blog. A caveat up front: I’m much more familiar with Eleventy than I am with Astro, which will likely be apparent when I get to the incremental build tests later in this post.

For Eleventy I used eleventy-netlify-boilerplate with zero modifications, and for Astro I used the blog template as described in their docs with some small modifications along the lines of this wordpress-to-astro repo to get categories and tags working. I didn’t want to use wordpress-to-astro directly since it was last updated two years ago, but it is a good reference point.

My blog has 770 posts which were exported to 770 Markdown files. With a paginated feed, categories, and tags, the total number of built pages is around 2550.*

Based on an average taken from 10 builds, Astro took 10.07 seconds and Eleventy took 4.29 seconds to build.

Incremental builds can speed things up significantly since the only built content is that which is relevant to the modified files.

Eleventy has supported incremental builds since December 2022 (I believe!), but it doesn’t yet support it on a CI server. There is an open issue for it which looks like it has traction.

To test incremental builds, I added and removed the same single tag on the same post 10 times.** Based on an average taken from 10 incremental builds, Eleventy took 2.17 seconds skipping 777 files. I would have expected it to skip more, but this might have to do with not being able to incrementally build paginated data.

I wanted to test the same content change in Astro… but it isn’t clear to me that there is an apples-to-apples comparison. Astro introduced an experimental Incremental Content Caching feature in v4.0 (not sure if this is supported on CI servers). When I added experimental.contentCollectionCache to the config, there was no difference between basic build times when I made a content change. I’m not sure if this is because having all of my content in Markdown makes the caching a mute point, or if it’s something else. If anyone has further context on how best to test Incremental Content Caching in Astro, would love to know.

For what it’s worth, running astro dev is extremely quick, just 125ms before it’s ready.

I’d be curious to do a similar benchmark using WordPress’s REST API but am not sure I’ll have the time… Will update here if I do.

* I give a rough number because the Eleventy boilerplate and Astro template generate a few additional pages, but the page total difference is in the single digits so I didn’t waste time evening them up perfectly.

** For my own future reference in case I do further tests: Add and remove the tag hello from this post.

Published by

“‘AI’ is pretty much just shorthand for mediocre”

Just read through “You sound like a bot” by Adi Robertson in the Verge. I hadn’t really put my finger on the right word for my feelings about AI until reading that article but that’s it: it feels very mediocre.

If you want to get a rough overview of how the average frontend engineer might feel about a JavaScript framework, ChatGPT is useful enough. If you’re willing to ignore the questionable origins of the training data in use, Midjourney can be useful for rapid image generation for an early storyboard.

But as of right now, the output always feels meh, “yeah ok”. Never really surprises you with a unique perspective, or an unexpected visual language. That vibe is only becoming stronger as AI developers continue to sand off the “rough” edges on their products.

Maybe that will change. As Robertson says, “Maybe the schism between artists and AI developers will resolve, and we’ll see more tools that amplify human idiosyncrasy instead of offering a lowest-common-denominator replacement for it.”

That’s not happening any time soon. One reason is that artists have been given about 1,000 reasons to distrust AI, and I think that it is only widespread artistic use and input that could actually lead to that sort of breakthrough.

Another reason: spewing mediocrity is a pretty strong sweet spot for AI. AI is useful as a summarizer so long as you take the response with a grain of salt and follow up on sources. Case in point: Elicit seems pretty cool! Listen to this ShopTalk Show episode with Maggie Appleton for more.

Anyways, maybe we’ll eventually get to the point where AI has that human “spark”, who knows. Maybe it’ll happen next month and I’ll eat my words. Until then, as most of the content we experience online becomes more grey and sludgy, the personal will become far more valuable.

In Anil Dash’s article “The Internet Is About to Get Weird Again” for Rolling Stone late last year, he says that “the human web, the one made by regular people, is resurgent”. He places a lot of emphasis on the breakdown of the content silos we’ve relied on for so many years, which definitely seems like the major catalyst for the shift. But AI’s growing mediocrity will be the force that drives it home and really makes the human web stick.

(Related side point: clearly I need to read Filterworld by Kyle Chayka.)

Edit 21 Feb 2024: Maybe I should eat my words sooner? OpenAI just came out with Sora. Which is impressive! But… IDK, it still feels meh somehow? Maybe it’s just because it’s still early days, we’ll see.