Turns out that young men kinda hate the things Trump is doing, and they can be organized quite effectively if we don't scold them.

the epistemological affect of a punctured elephant balloon that's on fire and hissing racisms

Sometime around 2016 everyone learned that “corrupt” was just some synonym of “bad” as a squishy rationalization for hating Hillary Clinton, and now here we are, showered in a repeat of an administration of endless actual corruption.

An underrated strategy for political power-building: Do things; tell people!

Get your students ready for a challenge! Registration is now open for the 2025-2026 High School Fed Challenge. Student teams from all over the nation will write researched podcast scripts on any topic related to the theme, “Economics of Music.” Learn more: www.federalreserve.gov/aboutthefed/...

Can someone smarter than me explain if/how Section 230 is relevant to this type of content that the platforms are, in fact, authoring and publishing? eiratansey.com/2025/12/20/i...

At some point I'll figure out a proper tagging+release strategy. Also the repo is now named ✨ agent-fec

 ✨ github.com/hodgesmr/age...

I try not to lean on the internet comparison too much, but recall we spent a decade shuffling adults into computer labs teaching them how to click a mouse.

More late-night shenanigans: OpenAI's Codex CLI now supports Agent Skills, so now you can use either Claude or Codex to crawl FEC files.

which 2028 Democrat is going to try to work a rehearsed “67” into a debate answer

People that just seem literally unable to conceptualize information about society as anything other than some kind of affirmation or attack on them and people they know personally.

The brave new world of a malicious Senate campaign sneaking prompt injections into the memo text field of their F99.

Late night baby shift also means late night weirdware. Tonight's edition: porting @dwillis.bsky.social's llm-fecfile plugin¹ (to @simonwillison.net's LLM CLI²) to Claude as an Agent Skill.³ ¹ github.com/dwillis/llm-... ² llm.datasette.io/en/stable/ ³ matthodges.com/posts/2025-1...

This is really interesting! But also why I hate generational cohort discourse. Every time you dig into it more, you decide you need more granular pigeon holes (Xennials?? Elder Millennials?? Young Gen Z??). It's almost like binning a continuous distribution into motivated-reasoning buckets is lossy!

I got tired of going to SpeechMeme.com to do a speech meme so I made a speech meme iOS shortcut. www.icloud.com/shortcuts/15...

I threw Claude into a Game Theory tournament. It built a Bayesian opponent model and beat some classic Iterated Prisoner’s Dilemma strategies. matthodges.com/posts/2025-1...

I deleted Slack off my phone. Its place on my Home Screen is now occupied by a diaper and feeding logging app. I love that. matthodges.com/posts/2025-1...

I wrote about why the marketing tech standard of hashing PII is silly, and I abused DuckDB into being a hashing mill. matthodges.com/posts/2025-1...

This week I achieved a few milestones: I hit 1,000 rides on Peloton, I weighed in at my goal, and I crossed five years of logging all this health data. I wrote a little bit about it here. matthodges.com/posts/2025-1...

VisiData rules. matthodges.com/posts/2025-0...

I’ve had Aristotle in my Goodreads queue for a decade. ChatGPT finally got me reading philosophy. matthodges.com/posts/2025-0...

I wrote about what Gödel, Escher, Bach teaches us about the impossibility of solving AI prompt injections. matthodges.com/posts/2025-0...

I wrote about what Gödel, Escher, Bach teaches us about the impossibility of solving AI prompt injections. matthodges.com/posts/2025-0...

I wrote about using LLM embeddings to normalize user data and how that could apply to political campaigns who want to better understand their supporters. No chatbots involved! matthodges.com/posts/2025-0...

did i ever tell you guys about the time i cracked open the DNC bracelet so that i could make a wearable strobe light controlled by my soundbar remote matthodges.com/posts/2024-0...

For years I've been throwing the same puzzle challenge at new GPT models. Every one has failed, until now. matthodges.com/posts/2025-0...