Blog · research
13 posts filed under research, newest first.
Stress on 12,000 Ukrainian words was a lookup, except for 246 arguments
Ukrainian never marks which syllable is stressed in print, so a learner's word list has to. I added the stress to all 12,000 words on my open Ukrainian frequency list: most came from an open dictio...
My best number was a word list read back to me
A month of work on the Ukrainian frequency list ended with a DOI, a live lemmatizer, a correction sent to the authors of the profile I lean on, and a headline result that dropped nine points once I...
Unratified
The next Invisible Infrastructure entry. The identity layer of digital public infrastructure does the four things a constitution does, with no ratifying moment. An explorable where you configure a ...
A signal that cannot lie
Conclave smoke, the presence dot, a porch light, firefly flashes, the CI badge, a whistling kettle, signals whose form is the message about a hidden state. A new interactive essay, two games, and a...
Making language models think faster.
Every token prediction wastes computation on impossible words. What if we only scored the valid ones? The math gets complicated fast.
What AI agents forget between conversations
Most AI systems lose track of context the moment you close the chat. SAGEN gives agents persistent memory of what's happening right now.
Special education as sorting infrastructure
What if racial disparities in special education aren't measurement errors? A new way to understand why some students get labeled disabled while others don't.
Schools are using broken lie detectors on students
Orion Newby wrote his own essay but got accused of cheating anyway. The AI that flagged him was wrong, but the damage was already done.
The invisible systems that sort your world.
Every database, form, and algorithm relies on categories. But who decides what counts? The answer reveals how power works in the information age.
Making AI Think Faster by Planning Ahead
When language models generate text, they waste enormous computational power. A new system learns when to take shortcuts and when to think hard.