@samuelstevens.bsky.social
So we built interactive demos where you can suppress specific features and watch model predictions change.
osu-nlp-group.github.io/SAE-V/#demos
See below for examples of what you can do.
osu-nlp-group.github.io/SAE-V/#demos
See below for examples of what you can do.
February 26, 2025 at 1:12 PM
So we built interactive demos where you can suppress specific features and watch model predictions change.
osu-nlp-group.github.io/SAE-V/#demos
See below for examples of what you can do.
osu-nlp-group.github.io/SAE-V/#demos
See below for examples of what you can do.
What's actually different between CLIP and DINOv2? CLIP knows what "Brazil" looks like: Rio's skyline, sidewalk patterns, and soccer jerseys.
We mapped 24,576 visual features in vision models using sparse autoencoders, revealing surprising differences in what they understand.
We mapped 24,576 visual features in vision models using sparse autoencoders, revealing surprising differences in what they understand.
February 26, 2025 at 1:12 PM
What's actually different between CLIP and DINOv2? CLIP knows what "Brazil" looks like: Rio's skyline, sidewalk patterns, and soccer jerseys.
We mapped 24,576 visual features in vision models using sparse autoencoders, revealing surprising differences in what they understand.
We mapped 24,576 visual features in vision models using sparse autoencoders, revealing surprising differences in what they understand.