Multimodal Model Diffing for Feature Discovery and Control | Paper Citation Record

Observation 63a68806-aeda-41aa-8974-f329e2ebba67 · outbound

This paper cites Pixtral 12b: A new frontier in image and text understanding.

Multimodal Model Diffing for Feature Discovery and Control Pixtral 12b: A new frontier in image and text understanding

Reference 1

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.734741Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation e980897a-63ad-4ab4-9cae-ab1bace9e2e9 · outbound

This paper cites Golden gate Claude.

Multimodal Model Diffing for Feature Discovery and Control Golden gate Claude

Reference 2

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.740439Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation b478f42b-d307-42c5-b16b-5954735aa5a1 · outbound

This paper cites SAE on activation differences.

Multimodal Model Diffing for Feature Discovery and Control SAE on activation differences

Reference 3

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.744964Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation eed6e3f5-9874-4c87-b480-11f0d9fdb4ad · outbound

This paper cites Refusal in Language Models Is Mediated by a Single Direction.

Multimodal Model Diffing for Feature Discovery and Control Refusal in Language Models Is Mediated by a Single Direction

Reference 4

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.749442Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 15f9a23f-fb05-4140-b13b-89fcd2bd4712 · outbound

This paper cites Revisiting model stitching to compare neural representations.Advances in neural information processing systems, 34:225–236, 2021.

Multimodal Model Diffing for Feature Discovery and Control Revisiting model stitching to compare neural representations.Advances in neural information processing systems, 34:225–236, 2021

Reference 5

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.754302Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 99bad67a-8f11-4798-a8d3-1e4ee9868bcc · outbound

This paper cites Representation Topology Divergence: A Method for Comparing Neural Network Representations.

Multimodal Model Diffing for Feature Discovery and Control Representation Topology Divergence: A Method for Comparing Neural Network Representations

Reference 6

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.758852Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation fb3300c3-6ac8-4870-bef9-0d7a1e819f96 · outbound

This paper cites Understanding information storage and transfer in multi-modal large language models.Advances in Neural Information Processing Systems, 37:7400–7426, 2024.

Multimodal Model Diffing for Feature Discovery and Control Understanding information storage and transfer in multi-modal large language models.Advances in Neural Information Processing Systems, 37:7400–7426, 2024

Reference 7

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.764254Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 775b35d7-5787-42c1-b28c-727175dec30e · outbound

This paper cites Towards monosemanticity: Decomposing language models with dictionary learning.Transformer Circuits Thread, 2023.

Multimodal Model Diffing for Feature Discovery and Control Towards monosemanticity: Decomposing language models with dictionary learning.Transformer Circuits Thread, 2023

Reference 8

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.768697Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation bfb2af8f-ac38-41fa-a848-228a761cca3a · outbound

This paper cites Stage-wise model diffing.

Multimodal Model Diffing for Feature Discovery and Control Stage-wise model diffing

Reference 9

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.773486Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 96afae2d-fab5-435d-bf48-229175c2211e · outbound

This paper cites Observing and controlling features in vision-language-action models.arXiv preprint arXiv:2603.05487, 2026.

Multimodal Model Diffing for Feature Discovery and Control Observing and controlling features in vision-language-action models.arXiv preprint arXiv:2603.05487, 2026

Reference 10

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.778192Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation bae5706b-c4c3-4713-829f-3cfd8479df19 · outbound

This paper cites Improving Steering Vectors by Targeting Sparse Autoencoder Features.

Multimodal Model Diffing for Feature Discovery and Control Improving Steering Vectors by Targeting Sparse Autoencoder Features

Reference 11

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.782849Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 21aef9b8-39d4-4a4a-bb5e-514a373a0a0d · outbound

This paper cites Pappas, Florian Tramer, Hamed Hassani, and Eric Wong.

Multimodal Model Diffing for Feature Discovery and Control Pappas, Florian Tramer, Hamed Hassani, and Eric Wong

Reference 12

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.788114Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 943068ae-9db0-4d97-a492-eea47ed36f53 · outbound

This paper cites Interpreting and Controlling Vision Foundation Models via Text Explanations.

Multimodal Model Diffing for Feature Discovery and Control Interpreting and Controlling Vision Foundation Models via Text Explanations

Reference 13

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.793237Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 729e283c-fe58-406d-a371-03d8e94288e4 · outbound

This paper cites LLaVA-MORE: A Comparative Study of LLMs and Visual Backbones for Enhanced Visual Instruction Tuning.

Multimodal Model Diffing for Feature Discovery and Control LLaVA-MORE: A Comparative Study of LLMs and Visual Backbones for Enhanced Visual Instruction Tuning

Reference 14

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.798668Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 47c9ed4e-1705-4432-9bf6-b809e4061e02 · outbound

This paper cites Explaining How Visual, Textual and Multimodal Encoders Share Concepts.

Multimodal Model Diffing for Feature Discovery and Control Explaining How Visual, Textual and Multimodal Encoders Share Concepts

Reference 15

Resolution: verified exact; local_arxiv, observed 2026-08-11T04:17:57.283192Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 9ea49434-9b1f-4ea9-b45e-0b22dc3bbf00 · outbound

This paper cites Sparse Autoencoders Find Highly Interpretable Features in Language Models.

Multimodal Model Diffing for Feature Discovery and Control Sparse Autoencoders Find Highly Interpretable Features in Language Models

Reference 16

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.809262Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ef347515-df97-4217-990c-bac9c5c5f962 · outbound

This paper cites Case study: Interpreting, manipulating, and controlling CLIP with sparse autoencoders.

Multimodal Model Diffing for Feature Discovery and Control Case study: Interpreting, manipulating, and controlling CLIP with sparse autoencoders

Reference 17

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.814349Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c03b7f1f-509a-432a-bf1a-565eab037242 · outbound

This paper cites Toy Models of Superposition.

Multimodal Model Diffing for Feature Discovery and Control Toy Models of Superposition

Reference 18

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.819230Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 907a2509-8501-4d4e-b1ce-4313ef05a681 · outbound

This paper cites Why does unsupervised pre-training help deep learning? 11:625–660, March.

Multimodal Model Diffing for Feature Discovery and Control Why does unsupervised pre-training help deep learning? 11:625–660, March

Reference 19

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.824530Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 5dc90ba2-6e4d-4bbf-98b4-ace19979d697 · outbound

This paper cites Interpreting CLIP's Image Representation via Text-Based Decomposition.

Multimodal Model Diffing for Feature Discovery and Control Interpreting CLIP's Image Representation via Text-Based Decomposition

Reference 20

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.829626Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 4227ee35-1ef9-4683-9e9f-b03e81994704 · outbound

This paper cites Scaling and evaluating sparse autoencoders.

Multimodal Model Diffing for Feature Discovery and Control Scaling and evaluating sparse autoencoders

Reference 21

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.834643Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation a67d0a03-cb2c-48bc-99b6-5077d2ca94a8 · outbound

This paper cites Gemma 2: Improving Open Language Models at a Practical Size.

Multimodal Model Diffing for Feature Discovery and Control Gemma 2: Improving Open Language Models at a Practical Size

Reference 22

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.839825Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 2891bc5a-3300-4926-b5a2-2e673fc4e1d2 · outbound

This paper cites FigStep: Jailbreaking large vision-language models via typographic visual prompts.

Multimodal Model Diffing for Feature Discovery and Control FigStep: Jailbreaking large vision-language models via typographic visual prompts

Reference 23

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.845475Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation eb7628dc-8244-471c-97d8-776e74cf4700 · outbound

This paper cites Making the v in vqa matter: Elevating the role of image understanding in visual question answering.

Multimodal Model Diffing for Feature Discovery and Control Making the v in vqa matter: Elevating the role of image understanding in visual question answering

Reference 24

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.850637Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 383f8bd4-0f3d-4f1e-8663-f5c8552e8985 · outbound

This paper cites Not all features are created equal: A mechanistic study of vision-language-action models.

Multimodal Model Diffing for Feature Discovery and Control Not all features are created equal: A mechanistic study of vision-language-action models

Reference 25

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.856252Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 594550ea-48d5-4b18-a375-73064b1a6c2e · outbound

This paper cites The Llama 3 Herd of Models.

Multimodal Model Diffing for Feature Discovery and Control The Llama 3 Herd of Models

Reference 26

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.861229Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 51b802d7-af4b-4cde-979e-97fc6d7c7e49 · outbound

This paper cites Mechanistic interpretability for steering vision-language-action models.

Multimodal Model Diffing for Feature Discovery and Control Mechanistic interpretability for steering vision-language-action models

Reference 27

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.865869Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3b640386-0590-4017-8b10-8dcab1ed9d2b · outbound

This paper cites Llama Scope: Extracting Millions of Features from Llama-3.1-8B with Sparse Autoencoders.

Multimodal Model Diffing for Feature Discovery and Control Llama Scope: Extracting Millions of Features from Llama-3.1-8B with Sparse Autoencoders

Reference 28

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.870806Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation e94666bd-7d15-4980-b0a4-2c789fe46bb3 · outbound

This paper cites In-context learning creates task vectors.

Multimodal Model Diffing for Feature Discovery and Control In-context learning creates task vectors

Reference 29

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.875585Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ed622c66-b2d4-4498-8016-67dc19488d7e · outbound

This paper cites VLSBench: Unveiling Visual Leakage in Multimodal Safety.

Multimodal Model Diffing for Feature Discovery and Control VLSBench: Unveiling Visual Leakage in Multimodal Safety

Reference 30

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.880064Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 92e3a5d3-924f-4d22-961c-de23b2368db0 · outbound

This paper cites Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training.

Multimodal Model Diffing for Feature Discovery and Control Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 31

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.885168Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 60bfe481-7159-4238-b912-6accdf089da1 · outbound

This paper cites Interpreting and Editing Vision-Language Representations to Mitigate Hallucinations.

Multimodal Model Diffing for Feature Discovery and Control Interpreting and Editing Vision-Language Representations to Mitigate Hallucinations

Reference 32

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.889674Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation dd219166-8d27-475a-a1c4-06ce11c66684 · outbound

This paper cites A “diff” tool for AI: Finding behavioral differences in new models.

Multimodal Model Diffing for Feature Discovery and Control A “diff” tool for AI: Finding behavioral differences in new models

Reference 33

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.894475Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 7bdee4c9-2165-49f5-a2c8-8a114e2a8644 · outbound

This paper cites Bridging the VLM and mech interp communities for multimodal interpretability.

Multimodal Model Diffing for Feature Discovery and Control Bridging the VLM and mech interp communities for multimodal interpretability

Reference 34

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.900408Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 0a1b3935-db61-4b7d-b663-d9f6cdf18b9d · outbound

This paper cites Steering CLIP's vision transformer with sparse autoencoders.

Multimodal Model Diffing for Feature Discovery and Control Steering CLIP's vision transformer with sparse autoencoders

Reference 35

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.905911Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 02d7cbe7-c662-4f20-a2f1-8f252dca6f28 · outbound

This paper cites Prisma: An Open Source Toolkit for Mechanistic Interpretability in Vision and Video.

Multimodal Model Diffing for Feature Discovery and Control Prisma: An Open Source Toolkit for Mechanistic Interpretability in Vision and Video

Reference 36

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.911157Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 4e56b9bf-2caf-4a64-8b63-4fffadd4fa48 · outbound

This paper cites Analyzing Finetuning Representation Shift for Multimodal LLMs Steering.

Multimodal Model Diffing for Feature Discovery and Control Analyzing Finetuning Representation Shift for Multimodal LLMs Steering

Reference 37

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.916191Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation bb729be5-e31f-4c6f-95d7-d9726ae0138e · outbound

This paper cites Saes (usually) transfer between base and chat models.

Multimodal Model Diffing for Feature Discovery and Control Saes (usually) transfer between base and chat models

Reference 38

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.921070Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 16f919f1-5867-4ed1-b1a8-49dea9800dbf · outbound

This paper cites Similarity of neural network representations revisited.

Multimodal Model Diffing for Feature Discovery and Control Similarity of neural network representations revisited

Reference 39

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.926358Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 07a5bec2-b757-4592-8d0b-4eccf85b6d89 · outbound

This paper cites Sakla, and Kowshik Thopalli.

Multimodal Model Diffing for Feature Discovery and Control Sakla, and Kowshik Thopalli

Reference 40

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.931559Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 62157067-b288-4ad7-97cb-c5b4e1633c98 · outbound

This paper cites Understanding image representations by measuring their equivariance and equivalence.

Multimodal Model Diffing for Feature Discovery and Control Understanding image representations by measuring their equivariance and equivalence

Reference 41

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.936739Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ef4f6fd3-9c05-4ffc-8be7-8a071882afe6 · outbound

This paper cites LLaVA-NeXT-Interleave: Tackling Multi-image, Video, and 3D in Large Multimodal Models.

Multimodal Model Diffing for Feature Discovery and Control LLaVA-NeXT-Interleave: Tackling Multi-image, Video, and 3D in Large Multimodal Models

Reference 42

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.941614Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ae34dab8-af93-4449-8e3a-061503ffd727 · outbound

This paper cites Inference- time intervention: Eliciting truthful answers from a language model.

Multimodal Model Diffing for Feature Discovery and Control Inference- time intervention: Eliciting truthful answers from a language model

Reference 43

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.946783Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f66078d7-2322-48be-8ac4-5c1c808c4191 · outbound

This paper cites Images are Achilles’ heel of alignment: Exploiting visual vulnerabilities for jailbreaking multimodal large language models.

Multimodal Model Diffing for Feature Discovery and Control Images are Achilles’ heel of alignment: Exploiting visual vulnerabilities for jailbreaking multimodal large language models

Reference 44

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.969271Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 7c5cf49e-69d0-46fc-b00c-d062e287ae2c · outbound

This paper cites Convergent Learning: Do different neural networks learn the same representations?.

Multimodal Model Diffing for Feature Discovery and Control Convergent Learning: Do different neural networks learn the same representations?

Reference 45

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.956223Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation b362ad44-4a2b-4a9d-9394-b8618f4cc42d · outbound

This paper cites Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2.

Multimodal Model Diffing for Feature Discovery and Control Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2

Reference 46

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.961255Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f5eddaa7-8d8b-40f1-9231-425afcca5f9e · outbound

This paper cites Sparse autoencoders reveal selective remapping of visual concepts during adaptation.

Multimodal Model Diffing for Feature Discovery and Control Sparse autoencoders reveal selective remapping of visual concepts during adaptation

Reference 47

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.967019Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 74cb08c5-3e38-4f66-b5a3-cf3e6d9061f1 · outbound

This paper cites A Survey on Mechanistic Interpretability for Multi-Modal Foundation Models.

Multimodal Model Diffing for Feature Discovery and Control A Survey on Mechanistic Interpretability for Multi-Modal Foundation Models

Reference 48

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.972218Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 50c39459-8359-4d8a-8f3b-06a5764fccdb · outbound

This paper cites Sparse crosscoders for cross-layer features and model diffing, October 25 2024.

Multimodal Model Diffing for Feature Discovery and Control Sparse crosscoders for cross-layer features and model diffing, October 25 2024

Reference 49

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.952495Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation e242e417-d6d1-466c-8c67-75d87eb17f48 · outbound

This paper cites Visual spatial reasoning.Transactions of the Association for Computational Linguistics, 2023.

Multimodal Model Diffing for Feature Discovery and Control Visual spatial reasoning.Transactions of the Association for Computational Linguistics, 2023

Reference 50

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.934987Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 428d81df-84f7-4297-95c6-84ba78dfca31 · outbound

This paper cites Visual instruction tuning.Advances in neural information processing systems, 36:34892–34916, 2023.

Multimodal Model Diffing for Feature Discovery and Control Visual instruction tuning.Advances in neural information processing systems, 36:34892–34916, 2023

Reference 51

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.986160Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 159ce224-eb30-4005-bd03-3e7034b33e31 · outbound

This paper cites Improved baselines with visual instruction tuning.

Multimodal Model Diffing for Feature Discovery and Control Improved baselines with visual instruction tuning

Reference 52

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.990284Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation a8f7a635-04af-41c1-9058-446d263e58fb · outbound

This paper cites MM-SafetyBench: A benchmark for safety evaluation of multimodal large language models.

Multimodal Model Diffing for Feature Discovery and Control MM-SafetyBench: A benchmark for safety evaluation of multimodal large language models

Reference 53

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.897139Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 4cc32df0-f6f3-4a06-9080-8ad98d0fa6c4 · outbound

This paper cites OCRBench: On the Hidden Mystery of OCR in Large Multimodal Models.

Multimodal Model Diffing for Feature Discovery and Control OCRBench: On the Hidden Mystery of OCR in Large Multimodal Models

Reference 54

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:55.999133Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 2db28554-8e14-41be-8bc8-fc9bd7333573 · outbound

This paper cites Michaud, Yonatan Belinkov, David Bau, and Aaron Mueller.

Multimodal Model Diffing for Feature Discovery and Control Michaud, Yonatan Belinkov, David Bau, and Aaron Mueller

Reference 55

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.880397Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 1012ade3-0d1d-4240-a17c-094795998f39 · outbound

This paper cites Locating and editing factual associations in GPT.

Multimodal Model Diffing for Feature Discovery and Control Locating and editing factual associations in GPT

Reference 56

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.008128Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 464d7b55-d456-457f-86c9-7b103017259e · outbound

This paper cites Robustly identifying concepts introduced during chat fine-tuning using crosscoders.arXiv preprint arXiv:2504.02922, 2025.

Multimodal Model Diffing for Feature Discovery and Control Robustly identifying concepts introduced during chat fine-tuning using crosscoders.arXiv preprint arXiv:2504.02922, 2025

Reference 57

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.012960Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation cc5bc455-3eba-42e8-901f-b572d332ca4a · outbound

This paper cites What we learned trying to diff base and chat models (and why it matters).LessWrong, 2025.

Multimodal Model Diffing for Feature Discovery and Control What we learned trying to diff base and chat models (and why it matters).LessWrong, 2025

Reference 58

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.853986Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 8451658b-066d-4558-848c-483cfff8b398 · outbound

This paper cites Insights on crosscoder model diffing.

Multimodal Model Diffing for Feature Discovery and Control Insights on crosscoder model diffing

Reference 59

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.837250Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 5db2cfa0-3178-48d0-9c2f-245259e280e8 · outbound

This paper cites Attribution patching: Activation patching at industrial scale.

Multimodal Model Diffing for Feature Discovery and Control Attribution patching: Activation patching at industrial scale

Reference 60

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.818307Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation cdd46c79-02f1-4c4e-86bb-b856ba020424 · outbound

This paper cites Towards Interpreting Visual Information Processing in Vision-Language Models.

Multimodal Model Diffing for Feature Discovery and Control Towards Interpreting Visual Information Processing in Vision-Language Models

Reference 61

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.032176Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 623a2370-84b8-4f4a-bf0b-0b8c4a7bd95c · outbound

This paper cites Steering Language Model Refusal with Sparse Autoencoders.

Multimodal Model Diffing for Feature Discovery and Control Steering Language Model Refusal with Sparse Autoencoders

Reference 62

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.038373Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation bbc722ed-96ce-4a69-89d5-624cf509d697 · outbound

This paper cites Zoom in: An introduction to circuits.Distill, 2020.

Multimodal Model Diffing for Feature Discovery and Control Zoom in: An introduction to circuits.Distill, 2020

Reference 63

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.044310Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 058d2038-410f-42cd-9c2b-798bf255d421 · outbound

This paper cites Visualizing representations: Deep learning and human beings.

Multimodal Model Diffing for Feature Discovery and Control Visualizing representations: Deep learning and human beings

Reference 64

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.798542Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 60c5603e-ea1d-40e2-b130-362652456e2b · outbound

This paper cites Probing the representational power of sparse autoencoders in vision models.

Multimodal Model Diffing for Feature Discovery and Control Probing the representational power of sparse autoencoders in vision models

Reference 65

Resolution: verified exact; raw_fallback, observed 2026-08-11T04:17:56.747243Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 5b9793e9-4c4a-4e7c-862e-920e5d638bbe · outbound

This paper cites Gpt-4o-mini: Advancing cost-efficient intelligence.

Multimodal Model Diffing for Feature Discovery and Control Gpt-4o-mini: Advancing cost-efficient intelligence

Reference 66

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.782692Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 377960dc-cb41-4e38-a7f4-d2932749b680 · outbound

This paper cites Sparse autoencoders learn monosemantic features in vision-language models.

Multimodal Model Diffing for Feature Discovery and Control Sparse autoencoders learn monosemantic features in vision-language models

Reference 67

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.063731Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 1c619101-97ef-42f1-aae8-e70c8d36057a · outbound

This paper cites Towards vision-language mechanistic interpretability: A causal tracing tool for blip.

Multimodal Model Diffing for Feature Discovery and Control Towards vision-language mechanistic interpretability: A causal tracing tool for blip

Reference 68

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.766065Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 94b2dead-18bc-4a07-af33-a10015ae3873 · outbound

This paper cites Beyond I'm Sorry, I Can't: Dissecting Large Language Model Refusal.

Multimodal Model Diffing for Feature Discovery and Control Beyond I'm Sorry, I Can't: Dissecting Large Language Model Refusal

Reference 69

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.073278Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation eacb6670-9287-4508-b7cc-10e65ab0b142 · outbound

This paper cites Visual adversarial examples jailbreak aligned large language models.

Multimodal Model Diffing for Feature Discovery and Control Visual adversarial examples jailbreak aligned large language models

Reference 70

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.750674Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation d5566a4c-bbc4-4029-82c9-0268f9e08351 · outbound

This paper cites Qwen-Scope: An open sparse autoencoder suite for the Qwen model family.

Multimodal Model Diffing for Feature Discovery and Control Qwen-Scope: An open sparse autoencoder suite for the Qwen model family

Reference 71

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.733675Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation fe447d69-874d-4cbd-b5a0-ea082f32f82e · outbound

This paper cites Learning transferable visual models from natural language supervision.

Multimodal Model Diffing for Feature Discovery and Control Learning transferable visual models from natural language supervision

Reference 72

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.088107Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation eadbd835-4e5f-45e3-8fea-de3c5e559122 · outbound

This paper cites Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders.

Multimodal Model Diffing for Feature Discovery and Control Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders

Reference 73

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.092721Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 4641de98-7e7b-4fc5-a4ab-6302312a8412 · outbound

This paper cites Steering Llama 2 via Contrastive Activation Addition.

Multimodal Model Diffing for Feature Discovery and Control Steering Llama 2 via Contrastive Activation Addition

Reference 74

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.097463Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 8c892395-c64d-4597-88bb-e01c606b2917 · outbound

This paper cites Multi- modal neurons in pretrained text-only transformers.

Multimodal Model Diffing for Feature Discovery and Control Multi- modal neurons in pretrained text-only transformers

Reference 75

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.705410Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation f298f62e-2df4-4676-8705-f70f9d84fe17 · outbound

This paper cites SteerVLM: Robust model control through lightweight activation steering for vision language models.

Multimodal Model Diffing for Feature Discovery and Control SteerVLM: Robust model control through lightweight activation steering for vision language models

Reference 76

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.688536Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 830f7044-5311-48c7-b4a4-e492efe747ce · outbound

This paper cites LVLM-Interpret: An Interpretability Tool for Large Vision-Language Models.

Multimodal Model Diffing for Feature Discovery and Control LVLM-Interpret: An Interpretability Tool for Large Vision-Language Models

Reference 77

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.111721Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ea2aabab-eee9-460d-ba2a-2b5538ff20e5 · outbound

This paper cites PaliGemma 2: A Family of Versatile VLMs for Transfer.

Multimodal Model Diffing for Feature Discovery and Control PaliGemma 2: A Family of Versatile VLMs for Transfer

Reference 78

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.116317Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 652c6242-0c83-45b3-84c8-a17a7d903b88 · outbound

This paper cites Daniel Freeman, Theodore R.

Multimodal Model Diffing for Feature Discovery and Control Daniel Freeman, Theodore R

Reference 79

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.122564Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ad666547-3921-415d-9f0d-0f41acd4ac5e · outbound

This paper cites Li, Arnab Sen Sharma, Aaron Mueller, Byron C.

Multimodal Model Diffing for Feature Discovery and Control Li, Arnab Sen Sharma, Aaron Mueller, Byron C

Reference 80

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.659875Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation d37558b3-dc39-493d-bf70-b84cb8eedef5 · outbound

This paper cites Eyes wide shut? exploring the visual shortcomings of multimodal llms.

Multimodal Model Diffing for Feature Discovery and Control Eyes wide shut? exploring the visual shortcomings of multimodal llms

Reference 81

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.132540Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 4f30b446-4ab4-4a1a-a778-10e1ba21a3ed · outbound

This paper cites Steering Language Models With Activation Engineering.

Multimodal Model Diffing for Feature Discovery and Control Steering Language Models With Activation Engineering

Reference 82

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.137372Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation a5e3ee40-f7fd-46f4-9da1-684f3115af2c · outbound

This paper cites Too late to recall: The two-hop problem in multimodal knowledge retrieval.

Multimodal Model Diffing for Feature Discovery and Control Too late to recall: The two-hop problem in multimodal knowledge retrieval

Reference 83

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.628662Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation e97f989c-c33a-4426-8979-5210adbd71a5 · outbound

This paper cites How Visual Representations Map to Language Feature Space in Multimodal LLMs.

Multimodal Model Diffing for Feature Discovery and Control How Visual Representations Map to Language Feature Space in Multimodal LLMs

Reference 84

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.147155Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9ddfa148-7c15-4249-9c65-72225f1a76bd · outbound

This paper cites Steering away from harm: An adaptive approach to defending vision language model against jailbreaks.

Multimodal Model Diffing for Feature Discovery and Control Steering away from harm: An adaptive approach to defending vision language model against jailbreaks

Reference 85

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.611779Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 3ad407f8-f160-46c8-9936-e40a5904103c · outbound

This paper cites InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency.

Multimodal Model Diffing for Feature Discovery and Control InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency

Reference 86

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.156829Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3b14a2e9-6053-4aa2-9b6a-279769fe4d46 · outbound

This paper cites AdaShield: Safeguarding multimodal large language models from structure-based attack via adaptive shield prompting.

Multimodal Model Diffing for Feature Discovery and Control AdaShield: Safeguarding multimodal large language models from structure-based attack via adaptive shield prompting

Reference 87

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.595098Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 47a8435a-2c41-4f4a-81e8-3c19919d4f69 · outbound

This paper cites LLaVA-CoT: Let Vision Language Models Reason Step-by-Step.

Multimodal Model Diffing for Feature Discovery and Control LLaVA-CoT: Let Vision Language Models Reason Step-by-Step

Reference 88

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.167527Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 95f2b71b-43ef-4498-9931-0584ebf2d855 · outbound

This paper cites Qwen3 Technical Report.

Multimodal Model Diffing for Feature Discovery and Control Qwen3 Technical Report

Reference 89

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.172320Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 581e9dcf-2c1f-4833-b068-0610fedbe8bf · outbound

This paper cites SafeSteer: Adaptive subspace steering for efficient jailbreak defense in vision-language models.arXiv preprint arXiv:2509.21400, 2025.

Multimodal Model Diffing for Feature Discovery and Control SafeSteer: Adaptive subspace steering for efficient jailbreak defense in vision-language models.arXiv preprint arXiv:2509.21400, 2025

Reference 90

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.177100Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c093c9d3-d5cc-4865-8131-636221f4691f · outbound

This paper cites Sigmoid loss for language image pre-training.

Multimodal Model Diffing for Feature Discovery and Control Sigmoid loss for language image pre-training

Reference 91

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.181619Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 112922d6-827d-42aa-a3d2-3aea70380e8f · outbound

This paper cites Towards Best Practices of Activation Patching in Language Models: Metrics and Methods.

Multimodal Model Diffing for Feature Discovery and Control Towards Best Practices of Activation Patching in Language Models: Metrics and Methods

Reference 92

Resolution: unresolved; no resolver link, observed 2026-08-11T04:17:56.186065Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f2b68228-1a9b-4451-b843-561f9a915118 · outbound

This paper cites Cross-modal information flow in multimodal large language models.

Multimodal Model Diffing for Feature Discovery and Control Cross-modal information flow in multimodal large language models

Reference 93

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.566959Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 2619e432-0229-4ce4-9384-2c2445bcae20 · outbound

This paper cites Multimodal situational safety.

Multimodal Model Diffing for Feature Discovery and Control Multimodal situational safety

Reference 94

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.549778Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation e7d95dd4-96b3-42a8-bfa2-b16e8a5e4b3e · outbound

This paper cites Relocated.

Multimodal Model Diffing for Feature Discovery and Control Relocated

Reference 95

Resolution: malformed identifier; raw_fallback, observed 2026-08-11T04:17:57.530047Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation dd40ce95-4dfb-4b0e-bb9f-d4bfd52a465d · outbound

This paper cites an unresolved cited work.

Multimodal Model Diffing for Feature Discovery and Control Unresolved cited work

Reference 96

Resolution: unresolved; raw_fallback, observed 2026-08-11T04:17:57.512239Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 7bbbfb87-1a52-4a94-8530-dc262d50d734 · outbound

This paper cites an unresolved cited work.

Multimodal Model Diffing for Feature Discovery and Control Unresolved cited work

Reference 97

Resolution: unresolved; raw_fallback, observed 2026-08-11T04:17:57.494613Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 418b9983-293a-4b05-83aa-d201d05c83b7 · outbound

This paper cites an unresolved cited work.

Multimodal Model Diffing for Feature Discovery and Control Unresolved cited work

Reference 98

Resolution: unresolved; raw_fallback, observed 2026-08-11T04:17:57.478120Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 4447cbf9-6ef0-4020-9860-8acf0b877876 · outbound

This paper cites this neuron activates for.

Multimodal Model Diffing for Feature Discovery and Control this neuron activates for

Reference 99

Resolution: verified fuzzy; raw_fallback, observed 2026-08-11T04:17:57.462367Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Coverage vector

Reference resolution

Outbound references

Pith citing papers