Yossi Gandelsman

I am a Member of Technical Staff at OpenAI, working on generative models and multimodal understanding.

Previously, I was an AI researcher at Reve.

I received my Ph.D. from UC Berkeley, where I was advised by Prof. Alexei Efros, and my M.Sc. from the Weizmann Institute of Science, where I was advised by Prof. Michal Irani.

During my Ph.D., I spent time as a Research Scientist at Meta, Adobe, and Transluce. Before starting my Ph.D., I was a member of the Perception Team at Google Research (now Google DeepMind).

Selected Publications

Please visit my google scholar for the full list of publications


Neuron Populations Exhibit Divergent Selectivity with Scale

Jailbreaking Vision-Language Models Through the Visual Modality

  • Aharon Azulay*, Jan Dubinski*, Zhuoyun Li*, Atharv Mittal* and Yossi Gandelsman
  • ICML 2026
  • paper | project page | code

The Unreasonable Effectiveness of Text Embedding Interpolation for Continuous Image Steering

In-Context Representation Hijacking

Vision Transformers Don't Need Trained Registers

Same Task, Different Circuits: Disentangling Modality-Specific Mechanisms in VLMs

Interpreting the Repeated Token Phenomenon in Large Language Models

  • Itay Yona, Jamie Hayes, Ilia Shumailov, Federico Barbero and Yossi Gandelsman
  • ICML 2025
  • paper | code

Interpreting and Editing Vision-Language Representations
to Mitigate Hallucinations

  • Nick Jiang*, Anish Kachinthaya*, Suzie Petryk† and Yossi Gandelsman†
  • ICLR 2025
  • paper | project page | code

Interpreting the Second-Order Effects of Neurons in CLIP

Interpreting the Weight Space of Customized Diffusion Models

The More You See in 2D, the More You Perceive in 3D

  • Xinyang Han*, Zelin Gao*, Angjoo Kanazawa, Shubham Goel† and Yossi Gandelsman†
  • CVPR 2024 [Spotlight]
  • paper | project page | code

Interpreting CLIP's Image Representation via
Text-Based Decomposition

Rosetta Neurons: Mining the Common Units in a Model Zoo

  • Amil Dravid*, Yossi Gandelsman*, Alexei A. Efros and Assaf Shocher
  • ICCV 2023
  • paper | project page | code

Test-Time Training with Masked Autoencoders

Visual Prompting via Image Inpainting

  • Amir Bar*, Yossi Gandelsman*, Trevor Darrell, Amir Globerson and Alexei A. Efros
  • NeurIPS 2022
  • paper | project page | code

MyStyle: A Personalized Generative Prior

  • Yotam Nitzan, Kfir Aberman, Qiurui He, Orly Liba, Michal Yarom, Yossi Gandelsman, Inbar Mosseri, Yael Pritch and Daniel Cohen-Or
  • SIGGRAPH Asia 2022 (journal track)
  • paper | project page | video

Deep ViT Features as Dense Visual Descriptors

  • Shir Amir, Yossi Gandelsman, Shai Bagon and Tali Dekel
  • ECCVW 2022 ["WIMF" Best Spotlight Presentation]
  • paper | project page | code

Explaining in Style:
Training a GAN to explain a classifier in StyleSpace

  • Oran Lang*, Yossi Gandelsman*, Michal Yarom*, Yoav Wald*, Gal Elidan, Avinatan Hassidim, William T. Freeman, Phillip Isola, Amir Globerson, Michal Irani and Inbar Mosseri
  • ICCV 2021
  • paper | project page | code | video | blog post

Semantic Pyramid for Image Generation

  • Assaf Shocher*, Yossi Gandelsman*, Inbar Mosseri, Michal Yarom, Michal Irani, William T. Freeman and Tali Dekel
  • CVPR 2020 [Oral]
  • paper | project page | video

Double-DIP: Unsupervised Image Decomposition via
Coupled Deep-Image-Priors