what is this ▸
code on github →
Vector strokes optimized with two losses by region.
Based on CLIPDraw.
Mask (me) — MSE loss to match photo pixels.
Non-mask (background) — CLIP loss to match text prompt.
I'm currently based in sf working at a startup working on ~intelligence~. If you're interested in program synthesis, we are hiring! I'm trying to answer the questions:
- How are people and natural intelligences so data and compute efficient?
- How do we use these insights to inform AI research?
- How do we ensure advances in AI = advances for the average person?
Before joining industry research, I was in the Computational Cognitive Science Lab at MIT where I also ran track/cross country.
Outside of research, I run a lot and am mediocre at many hobbies (bass, cooking, weight lifting, tending to plants).
If you are interested in Bayesian cognition, program synthesis, or inference-as-optimization, reach out!