Mechanistic Interpretability for Vision Workshop has officially begun #CVPR2025 ! 🚀 Join us at Grand C1 Hall for insightful perspectives on the state of interpretability in vision models by Tamar Rott Shaham.

Come join us at the Mechanistic Interpretability for Vision @ CVPR2025 workshop today at the #CVPR2025 in room C1! We have an amazing lineup of speakers 🔊 including David Bau Sonia trevordarrell Aleksander Madry Antonio Torralba and Michal Irani 🙌🏻

Yes indeed! Here are 3 pieces of advice by David Bau for PhD students interested in Mech Interp

Erasing Concepts from FLUX.1 models 🔥 ESD now supports FLUX models, and we've recently added SDXL support as well. Check out the code and share your experiments with us: github.com/rohitgandikota… (This is a long-awaited request from our users - thank you for your patience!)

We’re gearing up for the Mechanistic Interpretability in Vision Workshop 2026 👀✨ Who would you love to see as an invited speaker this year? Drop your suggestions below ⬇️

Mechanistic Interpretability for Vision @ CVPR2025 Rudy Gilman

Tutorial: How to Train a Concept Slider LoRA with AI Toolkit For this tutorial, I show you how to train a detail slider for Qwen Image that allows you to increase or decrease the amount of detail in an image by adjusting the LoRA strength. Links in 🧵

What a cool way to build and play with concept sliders! No code - just use the AIToolkit UI by Ostris! There is also a video tutorial on how to use the toolkit 👇

A first look at Adobe Research's work at #ICCV2025! This year, our research spans breakthroughs in image generation, 3D reconstruction, content authenticity, and more — advancing the science behind the next generation of creative technologies. Check out the blog post to learn

I’m hiring PhD students for 2026 TTIC. More details here: ttic.edu/studentapplica…

Ever wonder why AI image generators produce the specific results they do? In this work, Adobe researchers explore ways to map the visual knowledge hidden inside diffusion models – and use it to give users more control over the output. adobe.ly/47evqvD #ICCV2025

Adobe Research Adobe Rohit Gandikota Cool Work! 🔥🎉

📢 Today I begin my first semester as faculty in Computer Science at Tel Aviv University! Excited to start this new journey, and grateful to teach & research where my own journey began 🩵

How can a language model find the veggies in a menu? New pre-print where we investigate the internal mechanisms of LLMs when filtering on a list of options. Spoiler: turns out LLMs use strategies surprisingly similar to functional programming (think "filter" from python)! 🧵

Our paper "Vision Transformers Don't Need Trained Registers" will appear as a Spotlight at NeurIPS 2025! We uncover the mechanism behind high-norm tokens and attention sinks in ViTs, propose a training-free fix, and recently added an analytical model -- more on that below. ⬇️

A key challenge for interpretability agents is knowing when they’ve understood enough to stop experimenting. Our NeurIPS Conference paper introduces a self-reflective agent that measures the reliability of its own explanations and stops once its understanding of models has converged.

Self reflection within interpretability agents can nudge them to stop over-experimenting! Checkout this thread for more details 👇