More Hidden Treasures

Meta, Music

I just found a list that summarizes everything I possibly wanted circa 1997.

  • U2 CDs
  • 64 megs of memory
  • Awe 64 Sound card or USB speakers
  • Mark 6 alto saxophone
  • Stereo:
    • Radio turner and mixer
    • CD Changer
    • Turntable with stereo output
    • Double tape deck
  • Books: anything by Isaac Asimov
  • Cool mousepad
  • Flatbed color scanner
  • 3D video card
  • Computer games
    • Quake II
    • Starcraft
    • Command and Conquer II: Red Alert
  • Fast CD-ROM with digital support
  • Cool speakers for Dolby 4-channel surround sound
  • Big Monitor
  • Caller ID
  • USB Port

By the metric of this list, I’ve achieved almost 100% success. It’s remarkable how the material things we want can change so much over the years, and how material things in and of themselves can become so much less important. I wonder if I made a list now, how would I look at it in five years?

Menu Revamped

Design, Meta

Okay, per Tantek’s suggestion I’ve made the menu entirely a nested list, and adjusted the CSS accordingly. I had to work out a few kinks with the specificity of the contextual selectors, but after that it was a breeze. However this has brought to my attention how terribly ugly the CSS ridge effect is in Mozilla—really unattractive. It’s a shame because a lot of people seem to like it, but I’m exploring other options. Structure is groovy. Hell is a world without the style attribute.

Future of Email

Books, Web Standards

A group I’m a part of is preparing to form a number of “working groups” and each group may prepare a number of documents. The proposed format for these documents is plain ASCII text wrapped at 74 characters. It’s not the IETF, and on the whole it seemed like a rather restrictive format to develop documents in, an opinion which I’ve been trying my best to communicate. The discussion is still ongoing, but there was a brief tangent where several people misconstrued my argument as being one for HTML email, which is a totally separate beast.

Anyway it got me thinking about how HTML email is almost universally condemned among tech-savvy email groups. The problem, I think, is not technological but in fact human. HTML email has the potential to be clean, structured markup that can add a number of rich elements that there is no standard way to add in plain ASCII, such as emphasis, links, quotes, and in general represent things in a more meaningful way. The problem is generally not in the receiving client; I can’t think of a client with no HTML reading support (even Pine does some). Also the MIME standard allows and encourages a plain text equivalent of all rich content. It’s a problem, to put it into Spiderman terms, of great power and great responsibility. Someone very near and dear to my heart sends me email with garish background, text that varies between large and red or some purple script font, and any number of images speckled about. To me an ideal solution would be an email composer that enforced strict separation of style and content, and a client which allowed any CSS attributes to be toggled at will. Someday, perhaps?

Homework from WordCamp

AI, Open Source, WordPress

A few times during the WordCamp US keynote, I asked for a show of hands and got mostly blank stares back, especially on the AI questions. That’s not a criticism — it’s a reading list. If the future is going to be built on this stuff, the WordPress community should be fluent in it. Here’s the homework.

How models are actually trained

When I asked who knew about the reinforcement learning models go through, Robert looked out and said “I see no yeses.” Base models are trained to predict the next word across huge amounts of text; the friendly, helpful behavior comes after, in post-training, where human/AI feedback reinforces some responses over others. If you use these tools every day, understanding that pipeline changes how you prompt them and what you expect from them.

Start with Hugging Face’s Illustrating RLHF, a visual explainer. Chip Huyen’s RLHF deep dive is the best longer read, and Anthropic’s Constitutional AI paper shows where post-training went next. If you only have time for one thing, watch Andrej Karpathy’s Deep Dive into LLMs — pretraining, fine-tuning, reinforcement learning, and why models behave the way they do, all in one sitting.

Uncensored models

Almost nobody had tried an OBLITERATED model. Since refusal behavior is added in post-training, researchers found it can also be removed — it turns out to be, roughly, a single direction inside the model that you can subtract out. Maxime Labonne’s Uncensor any LLM with abliteration explains the technique, and the paper behind it, Refusal in Language Models Is Mediated by a Single Direction, is short and genuinely fascinating. Whatever you think of the applications, it’s worth understanding that open weights mean the community can modify these models, the same way the GPL means you can modify WordPress.

The open-weight wave

Only a few hands went up for Qwen 3.8. The open-weight models coming out of Chinese labs — Qwen, DeepSeek, Kimi, GLM — are at or near the frontier at a fraction of the cost, and the smaller ones run on a laptop. I demoed one from this MacBook on stage. Token costs have dropped about 20x, and that curve isn’t slowing down.

The easiest way to feel this for yourself is to run one locally tonight: install Ollama or LM Studio, pull a small Qwen model, and talk to an intelligence that lives entirely on your own machine, no cloud required. For a community built on the four freedoms, that should feel familiar — and exciting.

Read one, watch one, run one. Then bring what you learned to your next meetup.

It is really important to me that WordPress community members ride this AI wave.

Errata: I said Automattic has 32 or 33 data centers; the actual number is 29, all co-located.

There’s the smart publishers, and then there’s the ones going out of business. WIRED is one of the smart ones, and just launched an awesome redesign on WordPress. From their editor-in-chief:

Back in 1994 we launched Hotwired, the first site with original editorial content created for the web. It was a digital home for reporting on the future of science, business, design, and technology. You’ve come to trust us over the past two decades, but our growth online has sometimes come too quickly and with some pain. When I took over as editor in chief in 2012, WIRED had an archive of more than 100,000 stories. That’s good! But they were spread out over more than a dozen different databases, sections, and homepages tenuously connected by virtual duct tape and chewing gum. The cleanup process—onerous and without a shred of glamour—took almost 15 months. But finally, last year, our engineers rolled out a newly unified site architecture built atop a single streamlined WordPress installation. And you didn’t notice a hiccup. Maybe you saw that pages loaded a touch faster. Stories looked more WIRED.

The story of the engineering behind it from Kathleen Vignos is also cool:

The redesign gives us the third incarnation of our Curator application, which started years ago as a separate Groovy on Grails application maintained by a single Java developer. Curator once consumed articles from 35 different blogs for curation on our homepage. When we migrated our 17 active WordPress blogs into one WordPress install, we also rewrote Curator in Cake PHP to match our WordPress PHP backend. After this, anyone on our team could maintain Curator—but the architecture remained the same and lived outside of WordPress. Using this version of Curator, our web producer team manually constructed the homepage throughout each day as various stories were ready to be promoted.

Our new and improved Curator is now a custom WordPress plugin—and it’s artificially intelligent! This allows our homepage and section landing pages to be both automated and curated at the same time. Stories flow through automagically based on editorial criteria, but editors can take control of the flow by locking stories in certain slots in our card system. This means our homepage and section landing pages are constantly changing with new stories all day long.

Curator sounds cool, as does the coming “longform feature article builder.”