Understanding machine intelligence.
Look inside the black box.
I'm Eris. I research language models, build software, and take photographs. Exploring how intelligence works — and where it takes us.
I'm researching the systems that are changing the world, and building things to understand them.
I research large language models, with a focus on mechanistic interpretability and activation steering. Currently doing AI safety research as part of SPAR Spring 2026.
Explore research↗Instruct Vectors shows that a trained vector can make a base model behave like an instruction-tuned model, using very little data and no modification of its weights.
Read Instruct Vectors↗Software, experiments, and interactive tools. From language model interfaces to visualizations, small utilities, and projects that follow a question somewhere interesting.
Explore projects↗Photography, writing, reading, and games. A record of things I notice, things I make, and things I think are worth spending time with.
View photography↗
A base model can behave like an instruction-tuned model. No weight changes. Very little data. Just a learned vector.
Read the research ↗I'm Eris. A researcher, software developer, and photographer, trying to understand a world that refuses to stand still.
Get in touch ↗Studying computer science and psychology. Working on AI safety through SPAR. Between projects: books, games, and a camera.