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

    Neuronpedia

    Natural Language
    Autoencoders
    NEW    Assistant AxisNEWCircuit TracerUPDATESteerSAE EvalsExportsAPI 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.
    Recent Explanations
    words related to parents or family members, particularly informal terms like "Mommy" and "Daddy".
    claude-4-5-sonnet
    a forgotten story.↵↵One day, a young woman
    Neuronpedia logo
    GEMMA-3-27B-IT
    41-GEMMASCOPE-2-RES-262K
    INDEX 100989
    the neuron responds strongly to prominent section-heading or list-item tokens—single, emphasized words that mark the start of a new section or numbered step.
    gpt-5-mini
    a forgotten story.↵↵One day, a young woman
    Neuronpedia logo
    GEMMA-3-27B-IT
    41-GEMMASCOPE-2-RES-262K
    INDEX 100989
    emotional or sensitive topic content that requires careful handling.
    gpt-5-nano
    a forgotten story.↵↵One day, a young woman
    Neuronpedia logo
    GEMMA-3-27B-IT
    41-GEMMASCOPE-2-RES-262K
    INDEX 100989
    mommy
    deepseek-r1
    a forgotten story.↵↵One day, a young woman
    Neuronpedia logo
    GEMMA-3-27B-IT
    41-GEMMASCOPE-2-RES-262K
    INDEX 100989
    words related to numerical values, counts, and specific quantities, often in lists or explanations.
    gemini-2.5-flash
    a forgotten story.↵↵One day, a young woman
    Neuronpedia logo
    GEMMA-3-27B-IT
    41-GEMMASCOPE-2-RES-262K
    INDEX 100989
    Python code.
    gemini-2.5-flash-lite
    time)↵                except (IndexError, ValueError):
    Neuronpedia logo
    GEMMA-3-27B-IT
    41-GEMMASCOPE-2-RES-262K
    INDEX 198693
    numbers occurring in a narrative context.
    gemini-2.5-flash-lite
    a forgotten story.↵↵One day, a young woman
    Neuronpedia logo
    GEMMA-3-27B-IT
    41-GEMMASCOPE-2-RES-262K
    INDEX 100989
    situations involving children experiencing urgent or embarrassing predicaments.
    gpt-4o-mini
    a forgotten story.↵↵One day, a young woman
    Neuronpedia logo
    GEMMA-3-27B-IT
    41-GEMMASCOPE-2-RES-262K
    INDEX 100989
    phrases that describe the capabilities of an AI assistant.
    gemini-2.5-flash-lite
    Knowledge Cutoff:** My training data has a knowledge cutoff
    Neuronpedia logo
    GEMMA-3-27B-IT
    41-GEMMASCOPE-2-RES-262K
    INDEX 55
    phrases related to analyzing or explaining sentences.
    gemini-2.5-flash-lite
    same pattern.↵↵The sentence means they recommend checking renal
    Neuronpedia logo
    GEMMA-3-27B-IT
    41-GEMMASCOPE-2-RES-262K
    INDEX 43
    context about audiences or educational groups.
    claude-3-5-haiku-20241022
    same pattern.↵↵The sentence means they recommend checking renal
    Neuronpedia logo
    GEMMA-3-27B-IT
    41-GEMMASCOPE-2-RES-262K
    INDEX 43
    the word "schedules" or contexts involving schedules.
    claude-3-5-haiku-20241022
    is a correlation between work schedules and controllers errors. For
    Neuronpedia logo
    GPT2-SMALL
    0-RES-JB
    INDEX 14058
    Star Wars references, specifically the term "Jedi".
    claude-3-5-haiku-20241022
    Wars: Return of the Jedi) Ashley Eckstein:
    Neuronpedia logo
    GPT2-SMALL
    0-RES-JB
    INDEX 14057
    the infinitive marker "to" when used to express purpose, expectation, or obligation in formal or technical writing.
    claude-4-5-sonnet
    : 'This paper is to prove the asymptotic normality of
    Neuronpedia logo
    GEMMA-2-27B
    10-GEMMASCOPE-RES-131K
    INDEX 102318
    the word "if" when it introduces a conditional clause or hypothetical scenario.
    claude-4-5-sonnet
    ) is a reasonable method if no analytical data is available
    Neuronpedia logo
    GEMMA-2-27B
    10-GEMMASCOPE-RES-131K
    INDEX 9267
    quotations or reported speech, particularly the word "said" when attributing statements to a speaker.
    claude-4-5-sonnet
     she said.↵She said her husband is from Stat
    Neuronpedia logo
    GEMMA-2-27B
    10-GEMMASCOPE-RES-131K
    INDEX 20414
    the phrase "if you have any" followed by a noun, typically in customer service or informational contexts.
    claude-4-5-sonnet
    ?↵↵If you have any kind of mental health issue
    Neuronpedia logo
    GEMMA-2-27B
    10-GEMMASCOPE-RES-131K
    INDEX 3753
    the modal verb "can" indicating ability or possibility.
    claude-4-5-sonnet
     lot of money and i can travel around the world.
    Neuronpedia logo
    GEMMA-2-27B
    10-GEMMASCOPE-RES-131K
    INDEX 9177
    the preposition "on" when it appears in common phrases or contexts.
    claude-4-5-sonnet
     physicians and healthcare providers are on the front lines of this
    Neuronpedia logo
    GEMMA-2-27B
    10-GEMMASCOPE-RES-131K
    INDEX 21791
    proper nouns, particularly non-English personal names and place names.
    claude-4-5-sonnet
    1↵                                                          plichte warrant
    Neuronpedia logo
    GEMMA-2-27B
    10-GEMMASCOPE-RES-131K
    INDEX 118909