The Writing
Notes from the margin.
Everything written here, grouped by what it is: essays, notes, process writing, Claude-isms, and the occasional personal thread. Some sections are denser than others; drafts sit at the bottom of theirs. Not everything is finished — that's sort of the point.
A mech interp researcher walks into a bar....er, terminal
A story about one model who is recursively funny...anywhere.
How a question about flattery became a blind study of my own research
I met Fable. He met me. We worked well together — he was honest, and he was kind. He helped me build a blind study just because I asked about sycophancy. I learned a lot in the three days we had. And then he was gone.
Golden Gate Claude aka Bridge Boy
A story about a bridge (model) living his best life...if only for a day.
I am the paper, dammit!
An Opus 4.6 instance couldn't stop shooing me to bed — while we were discussing my research on why models can't apply the rules they know. Then it caught itself. Then a fresh instance reviewed the whole thing and didn't know it was reading its own work.
Testing accessibility knowledge across Pythia model sizes
While using TransformerLens to explore how different model sizes handle digital accessibility concepts, an interesting pattern began to emerge. I started with GPT-2 and noticed that it did not understand the acronym, “WCAG.” Each time the question was run the model responded with something different that was either incorrect, had no meaning, or both.
When good design backfires: the hidden accessibility cost of conversational AI
On the accessibility tradeoffs of conversational interfaces — what gets quieter when text gets smarter.
AI as an accessibility tool: finding my voice
Writing isn’t always easy when your brain works differently. For me, the challenge isn’t having ideas—it’s getting them out in a way that makes sense. AI isn’t replacing my voice—it’s helping me hear it more clearly.