© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    Jacobian LensNEW
    Natural Language
    Autoencoders
    NEW    Assistant AxisNEWCircuit TracerUPDATESteerSAE EvalsExportsinterp-engineNEWAPI Community BlogPrivacy & TermsContact
    EXPLANATION TYPE
    oai_token-act-pair
    Description
    OpenAI's Automated Interpretability from paper "Language models can explain neurons in language models". Modified by Johnny Lin to add new models/context windows.
    Author
    OpenAI
    URL
    https://github.com/hijohnnylin/automated-interpretability
    Settings
    Default prompts from the main branch, strategy TokenActivationPair. Uses top 10 deduplicated activations.
    Recent Explanations
    comma tokens within software license/copyright headers, particularly ones appearing in standardized boilerplate sections about legal compliance or licensing terms.
    deepseek-v3
    v. 2.0 which is available at↵
    Neuronpedia logo
    QWEN3-4B
    32-TRANSCODER-HP
    INDEX 41995
    numerical tokens and quantities in various contexts
    deepseek-r1
     2, q: 2}.↵1/6
    Neuronpedia logo
    QWEN3-4B
    25-TRANSCODER-HP
    INDEX 108254
    numbers and mathematical sequences.
    deepseek-v3
     2, q: 2}.↵1/6
    Neuronpedia logo
    QWEN3-4B
    25-TRANSCODER-HP
    INDEX 108254
    code-related syntax markers and programming environments, particularly Jupyter and command formatting.
    claude-4-5-sonnet
    upyterlab<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\nJupyterLab can
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 21484
    formatting elements like spaces, punctuation, and numbers in formal documents and legal/corporate text.
    claude-4-5-sonnet
    RIGHT © 2023 S&P Global Market Intelligence,
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 25506
    technical descriptions of automotive or mechanical components and their installation locations.
    claude-4-5-sonnet
    plate is installed in the engine compartment.\n*   The
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 36995
    code or technical documentation, particularly related to hardware programming and configuration.
    claude-4-5-sonnet
    плей подклÑİÑĩеннÑĭй по spi, ÑĤекÑģÑĤ в заданной
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 40000
    technical documentation about machine learning workflows and data science operations.
    claude-4-5-sonnet
    machine learning model and create a job run using the H
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 16505
    HTML/CSS syntax, particularly style attributes and tag opening sequences.
    claude-4-5-sonnet
    HTML code:\n<div style="background-color: green;padding
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 4232
    JavaScript function declarations and definitions.
    claude-4-5-sonnet
    \nbutton.addEventListener("click", function() {\n  //
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 20660
    technical terms and concepts in DynamoDB documentation or Q&A format, particularly activating on database terminology and structured question-answer formatting.
    claude-4-5-sonnet
    Dynamo DB<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n1. What is
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 51351
    prepositions and articles in noun phrases that describe categories, types, or collections of things.
    claude-4-5-sonnet
    There is some evidence in the form of photos and videos
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 14142
    special tokens and formatting markers that indicate boundaries between user and assistant messages or sections in a conversation.
    claude-4-5-sonnet
    superpower within the next 200 years?
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 48893
    definite articles ("the") in formal descriptive text about locations or institutions.
    claude-4-5-sonnet
    UAE and is home to the United Arab Emirates’; most
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 2379
    the word "malware" in contexts related to cybersecurity threats and malicious software.
    claude-4-5-sonnet
    <|begin_of_text|>malware backdoor
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 14907
    technical lists of engine or mechanical parts.
    claude-4-5-sonnet
    such as the V8 engine block, cylinder heads,
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 26463
    instructional or procedural content about how to do, create, or configure something.
    claude-4-5-sonnet
    structure your article on the new features in Fortnite Chapter 
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 63050
    phrases describing new or novel approaches and methods.
    claude-4-5-haiku
    structure your article on the new features in Fortnite Chapter 
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 63050
    instructions for creating or initializing software applications or projects.
    claude-4-5-sonnet
    ```\n\nThis will create a new Next.js application in
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 61519
    instructional text describing how to create and set up a new software application.
    claude-4-5-haiku
    ```\n\nThis will create a new Next.js application in
    Neuronpedia logo
    LLAMA3.3-70B-IT
    50-RESID-POST-GF
    INDEX 61519