Image de couverture de livre

Towards eliciting latent knowledge from LLMs with mechanistic interpretability

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Cywiński, Bartosz, Ryd, Emil, Rajamanoharan, Senthooran, Nanda, Neel
Format:	Preprint
Publié:	2025
Sujets:	Machine Learning
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Eliciting Secret Knowledge from Language Models
par: Cywiński, Bartosz, et autres
Publié: (2025)

Convergent Linear Representations of Emergent Misalignment
par: Soligo, Anna, et autres
Publié: (2025)

Do I Know This Entity? Knowledge Awareness and Hallucinations in Language Models
par: Ferrando, Javier, et autres
Publié: (2024)

Thought Branches: Interpreting LLM Reasoning Requires Resampling
par: Macar, Uzay, et autres
Publié: (2025)

Model Organisms for Emergent Misalignment
par: Turner, Edward, et autres
Publié: (2025)

Are Sparse Autoencoders Useful? A Case Study in Sparse Probing
par: Kantamneni, Subhash, et autres
Publié: (2025)

Simple Mechanistic Explanations for Out-Of-Context Reasoning
par: Wang, Atticus, et autres
Publié: (2025)

Steering Out-of-Distribution Generalization with Concept Ablation Fine-Tuning
par: Casademunt, Helena, et autres
Publié: (2025)

Chain-of-Thought Reasoning In The Wild Is Not Always Faithful
par: Arcuschin, Iván, et autres
Publié: (2025)

Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders
par: Rajamanoharan, Senthooran, et autres
Publié: (2024)

How Well Do Models Follow Their Constitutions?
par: Jakkli, Arya, et autres
Publié: (2026)

Censored LLMs as a Natural Testbed for Secret Knowledge Elicitation
par: Casademunt, Helena, et autres
Publié: (2026)

Simple LLM Baselines are Competitive for Model Diffing
par: Kempf, Elias, et autres
Publié: (2026)

Improving Dictionary Learning with Gated Sparse Autoencoders
par: Rajamanoharan, Senthooran, et autres
Publié: (2024)

How to use and interpret activation patching
par: Heimersheim, Stefan, et autres
Publié: (2024)

Emergent Misalignment is Easy, Narrow Misalignment is Hard
par: Soligo, Anna, et autres
Publié: (2026)

Fine Flood Forecasts: Incorporating local data into global models through fine-tuning
par: Ryd, Emil, et autres
Publié: (2025)

SAeUron: Interpretable Concept Unlearning in Diffusion Models with Sparse Autoencoders
par: Cywiński, Bartosz, et autres
Publié: (2025)

Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2
par: Lieberum, Tom, et autres
Publié: (2024)

Removing Sandbagging in LLMs by Training with Weak Supervision
par: Ryd, Emil, et autres
Publié: (2026)

Subliminal Learning Is Steering Vector Distillation
par: Blank, Camila, et autres
Publié: (2026)

Towards Principled Evaluations of Sparse Autoencoders for Interpretability and Control
par: Makelov, Aleksandar, et autres
Publié: (2024)

Towards Best Practices of Activation Patching in Language Models: Metrics and Methods
par: Zhang, Fred, et autres
Publié: (2023)

Summing Up the Facts: Additive Mechanisms Behind Factual Recall in LLMs
par: Chughtai, Bilal, et autres
Publié: (2024)

GUIDE: Guidance-based Incremental Learning with Diffusion Models
par: Cywiński, Bartosz, et autres
Publié: (2024)

Dense SAE Latents Are Features, Not Bugs
par: Sun, Xiaoqing, et autres
Publié: (2025)

Bilinear MLPs enable weight-based mechanistic interpretability
par: Pearce, Michael T., et autres
Publié: (2024)

An introduction to graphical tensor notation for mechanistic interpretability
par: Taylor, Jordan K.
Publié: (2024)

Explorations of Self-Repair in Language Models
par: Rushing, Cody, et autres
Publié: (2024)

A mechanistically interpretable neural network for regulatory genomics
par: Tseng, Alex M., et autres
Publié: (2024)

Difficulties with Evaluating a Deception Detector for AIs
par: Smith, Lewis, et autres
Publié: (2025)

Opening the AI black box: program synthesis via mechanistic interpretability
par: Michaud, Eric J., et autres
Publié: (2024)

Inference-Time Decomposition of Activations (ITDA): A Scalable Approach to Interpreting Large Language Models
par: Leask, Patrick, et autres
Publié: (2025)

BatchTopK Sparse Autoencoders
par: Bussmann, Bart, et autres
Publié: (2024)

Transcoders Find Interpretable LLM Feature Circuits
par: Dunefsky, Jacob, et autres
Publié: (2024)

Inoculation Prompting: Instructing LLMs to misbehave at train-time improves test-time alignment
par: Wichers, Nevan, et autres
Publié: (2025)

How Visual Representations Map to Language Feature Space in Multimodal LLMs
par: Venhoff, Constantin, et autres
Publié: (2025)

What's the plan? Metrics for implicit planning in LLMs and their application to rhyme generation and question answering
par: Maar, Jim, et autres
Publié: (2026)

Reasoning-Finetuning Repurposes Latent Representations in Base Models
par: Ward, Jake, et autres
Publié: (2025)

RelP: Faithful and Efficient Circuit Discovery in Language Models via Relevance Patching
par: Jafari, Farnoush Rezaei, et autres
Publié: (2025)