DinoDreamer: browser tool visualizes DINOv3 representations with cellular automata
zzznah · x · 2026-11-21
Omar Hammami released DinoDreamer, an interactive interpretability tool that runs entirely in the browser (phones included) to explore Meta's self-supervised vision model DINOv3:
- A block-sparse featurizer maps DINOv3 representations into 16,384 basis groups, visualized via neural cellular automata
- You can browse every group, train an automaton to grow what it responds to, and disturb it by drawing — DeepDream-like, but for self-supervised vision models
- Motivation: interpretability should be a public service like data visualization was for big data, not just researcher-only "neuron found, behavior off" stories
- The writeup honestly covers the author's battles with overcomplete sparse dictionaries and where he thinks the method falls short
DINOv3 was trained on 1.7B unlabeled images via crop-consistency self-distillation.
More from Research
- Riemann hypothesis 'proven by Turkish mathematician using AI' claim mocked as the 'Organic Turk' — CatAstro_Piyush · 2026-11-22
- Andriy Burkov releases The Hundred-Page Language Models Book, hands-on with PyTorch — burkov · 2026-11-22
- Generative AI meets cardiology: Dagstuhl seminar bridges ML and clinical practice — StephanMandt · 2026-11-22
- SubRoformer hits SOTA music source separation with DSP subband filters — affige_yang · 2026-11-22
- Emily Riehl releases Lean game teaching dependent type theory at intro-to-proofs level — abhi9u · 2026-11-22
- Curated reading list: evaluating mobile-use agents on Android, from AppAgent to AndroidReality — East-Muffin-6472 · 2026-11-22