A Moist Owlette is freshly showered at Owl Park San Vigilio, a zoological park in northern Italy 🇮🇹 via https://fb.com/816647270688824 #owlsintowels 💛🦉

Ok, I know promised a thread on the fascinating and important new research explaining what happened with the rare, serious side effects of AstraZeneca's and Johnson&Johnson's Covid-19 vaccines.

So, Science story by my colleague Gretchen Vogel and me is here and 🧪🧵 is coming:
https://www.science.org/content/article/rare-dangerous-side-effects-some-covid-19-vaccines-explained

New: Researchers found a clever way to reveal AI biases, like the states with the laziest people, or the cities with the best pizza.

Look up your city here -> https://wapo.st/4bMoWqk with @geoffreyfowler.bsky.social

We know much more about the tragic school shooting in Tumbler Ridge.

Here is the latest, with information on the victims, the guns used, the assailant and how parents can get help.

https://thetyee.ca/News/2026/02/11/Tumbler-Ridge-Tragedy-What-We-Know/?utm_source=mastodon&utm_medium=social&utm_campaign=editorial

Allow me to recommend the Soviet production of Sherlock Holmes on YouTube. As an index of its vibe, this is how it depicts a man who has just arrived from Canada.

I hate this new sundial technology

https://mathewingram.com/work/2026/02/10/i-hate-this-new-sundial-technology/?utm_source=mastodon&utm_medium=jetpack_social

norway viking ship, around the world in 18 holes, lake george, new york, 2002

Must-read: How ‘Pink Slime’ Publishers Are Weaponizing FOIA

From Mirada Green and the Tow Center for Digital Journalism:

"Metric Media filed more than nine thousand public records requests last year. It used the data to target Democratic politicians and private citizens."

If you're unfamiliar with Metric Media, they own thousands of local "news" sites that mostly republish drivel until election time rolls around they're all partisan conservative publications masquerading as local news.

"Founded in 2019, Metric has been criticized for jury tampering and tied to pay-for-play political schemes and fake newspapers that land in mailboxes ahead of key elections. Recently, it has focused on obtaining troves of public records. In the past year, an investigation by the Tow Center for Digital Journalism has found Metric filed more than nine thousand Freedom of Information Act requests across all fifty states."

"Many of the requests are for data at the forefront of America’s culture-wars, from allegedly rigged elections to banned books to transgender inmates. With public records in hand, Metric has targeted liberal politicians with negative reporting, criticized the funding of nonprofit organizations, and published personally identifying details about small-town residents who spoke up at school board hearings. Unlike traditional journalism, Metric’s stories do not air dueling perspectives or offer targets a chance to comment."

https://www.cjr.org/tow_center/pink-slime-networks-are-weaponizing-foia.php

Taylor Rehmet's win in a Texas state senate shocked pundits.

But, as I found in interviews with people close to the ground, it's not so surprising.

A lot of it is about the books bans. People hate the book bans. Even some Republicans! An analysis.

https://www.salon.com/2026/02/06/shock-democratic-upset-in-texas-shows-voters-still-hate-book-bans/

One of the richest men in the world is unable to afford running a good newspaper.

"The Metro section will be restructured, ensuring a "healthy presence for local subscribers," he said. According to a Metro staffer who was just laid off, there will be about a dozen people left on the desk. That's down from more than 40."

https://www.npr.org/2026/02/04/nx-s1-5699328/washington-post-layoffs-jobs-bezos

Washington City Paper: Every Scare in Paranormal Activity Is Masterfully Earned

I saw this production last Friday and really enjoyed it!

"As the couple acclimates to their new life, more disturbing things start to occur. The jump scares start early but intensify as the play goes on: a randomly activated security alarm or a blaring ringtone early on tactfully build the anticipation for the more gruesome reveals of the second act. While Davis (who pulls double-duty as costume designer) crafts the pitch-perfect backdrop for these eerie events, Anna Watson’s lighting design is masterfully suspenseful. The cottage atmosphere is always punctuated with chill-inducing shadows and otherworldly flickering."

https://washingtoncitypaper.com/article/778550/every-scare-in-paranormal-activity-is-masterfully-earned/

#WashingtonDC #theatre #horror

Well, after 14 years, my Write a Compiler course is having its final curtain call February 16-20. You can still get in on the action though http://dabeaz.com/compiler.html.

Honestly, it's hard for me to believe that taking a class like this out of a university setting would even work for as long as it did. I definitely had a blast teaching it! But, alas, all good things eventually come to an end. Time to rotate the crops.

OK, so somehow they knew exactly how to get me 100% back on board with #MST3K. The description even says "The entire production will take place this spring/summer in Minnesota… We'll be building all-new, hand-crafted sets and props.”

Put it directly into my veins. https://www.kickstarter.com/projects/rifftrax/rifftrax-makes-mst3k

Often the agent gets into trouble when the unit tests it just wrote don't work as expected. When this occurs, the agent reacts much as I do: it starts adding print() calls to debug what's happening, printing out fragments of XML, or writing little test programs that call an API to check what happens.

So my little experiment worked nicely, for me. The quotation-tools package has been updated in the ways that I wanted, and now I can use it to make my quote-posting bot more featureful.

My next learning project will be something from scratch. I've decided to do an implementation of the logic for the board game Spots (https://boardgamegeek.com/boardgame/372559/spots) and then I'll put a basic TUI or other interface on top of it.

(4/4)

Armin Ronacher recently wrote a blogpost where he wonders "Is this where we are going? Am I just not ready for this new world? Are we all collectively getting insane?" https://lucumr.pocoo.org/2026/1/18/agent-psychosis/

And I get where Armin is coming from; frankly, Claude Code Guy may already be insane. It feels like you're the Sorceror's Appentice, where you can casually suggest "rewrite everything to use a different library" or "rearrange all of these methods", and the agent will just crank away and do it. Every whim can be carried out for you without pushback or complaint, which makes changes easy? SO EASY.

But I also noticed that I'm getting sloppy. The agent wrote lots of unit tests for the classes that I scanned through, but I didn't carefully consider each unit test as if I was writing it myself: what is this test exercising? Is it checking the right post-conditions? The code comes so easily that slowing down to read unit tests for a half hour felt like an interminable delay.

For the multi-page HTML output, I asked it to use good CSS and it chose a CSS framework called PicoCSS (https://picocss.com/). I skimmed the website to check that its features were what I wanted and that the license was acceptable, but I didn't go deep enough to know if the resulting code is using PicoCSS correctly. If I wanted to work in OCaml or Rust or some other language that I'm less familiar with, it would be really easy for me to get out over my skis.

(3/4)

Random thoughts:

Claude encourages formulating an implementation plan before letting the agent go wild. I enjoyed that process, as it forced thinking about what I wanted and how the change would need to be structured. Sometimes I'd just hand-edit the plan to correct it, or I'd talk the AI into making changes until I was satisfied.

You do need to watch what the agent is doing, though, as sometimes it makes inexplicable decisions. For example, there's a Quotation class that represents a quote and has to_text(), to_wiki(), to_html(), etc. methods to output the text in different formats. But when I asked the agent to produce a multi-page HTML output, it decided to use a regex to strip out HTML tags even though conversion to HTML already existed in the code. And that regex would have been a second QEL-to-text pathway, even though we already had a to_text() method, so it's a doubled mistake. Claude would have happily cranked away at this solution for hours even though it's ultimately unsuitable.

Wired recently ran an interview with the head of Claude Code, where he's quoted: "This morning I woke up, I think I started three or four coding agents on my phone. When I get into work, I'll check in on what they're doing, and then I'll probably start a few more in my terminal. At any given moment, I usually have like five or ten running across the terminal, mobile, and website." https://www.wired.com/story/claude-code-success-anthropic-business-model/

This seems like madness to me. Is he carefully proofreading all of the code produced by those three or four agents? Are problems avoidable with a detailed-enough plan, or are you always risking that the agent will go skidding off down a mistaken path, assembling a house of cards that just happens to work?

(2/4)

A few weekends ago I wanted to experiment with using Claude Code. It's the anointed AI-coding tool for use at work, so it'd be valuable for me to get more familiar with using it.

I used my quotation-tools Python project as my test bed: https://codeberg.org/akuchling/quotation-tools/

It parses files of quotations written in an XML DTD called Quotation Exchange Language, and can output them in various formats. About 9 months in an earlier experiment with Gemini, I made a branch that switched from the Python stdlib's pulldom module to using lxml. I had a bunch of other updates I wanted to make, such as modernizing the installation & setup code in general, adding a JSON output format to simplify implementation of my Mastodon-posted bots, and fixing an irritating bug in the handling of line breaks.

In many ways it's an ideal project for this experiment; the project only works with text input and output, doesn't require anything external such as a database or web server, and even a sizable test suite runs in a few seconds. I contrast it with our test suites at work, which take minutes to run and need complicated database servers to do anything.

(1/4)

#Claude #ArtificialIntelligence

Caught the travelling stage version of #ParanormalActivity at the Shakespeare Theatre in DC tonight.

I didn't like the original movie because the ending was completely unsatisfying for me. But I enjoyed this presentation, which I think is a new story with no connection. There are some onstage effects and jump scares that are really effective, the actors did a great job.

https://shakespearetheatre.org/index.php/events/paranormal-activity-25-26

#horror #theater

@BrettCoulstock @ClassicDWQuotes I'll check the DVD subtitles, as it's certainly possible I misheard. Though 'wet' gives an unpleasant physicality to the image, which is what struck me about it .

MeeplePhD: Top 10 New-To-Me Games Played in 2025 (17 min)

https://youtu.be/oSLqIo6OPao

#BoardGames #MeeplePhD