Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    APIAssistant AxisNEWCircuit TracerNEWSteerSAE EvalsExports Community BlogPrivacy & TermsContact
    © Neuronpedia 2025
    Privacy & TermsBlogGitHubSlackTwitterContact
    EXPLANATION TYPE
    np_max-act
    Description
    Forces concise explanations and shows the model the top activating tokens and texts. A simpler version of np_max-act-logits.
    Author
    Neuronpedia
    URL
    https://github.com/hijohnnylin/automated-interpretability/blob/917b11e38111c43526fe03ae6094a7081aeb982a/neuron_explainer/explanations/explainer.py#L1181
    Settings
    Activations shown = 24 tokens around max act. Shows model the max activating token too.
    Recent Explanations
    pipe-separated sentiment labels
    gpt-5
    Technical breakout↵10. Economic recovery↵↵Events before a
    Neuronpedia logo
    LLAMA3.1-8B-IT
    19-RESID-POST-AA
    INDEX 59638
    Medium publishing platform
    gemini-2.5-flash
    ([https://towardsdatascience.com/](https
    Neuronpedia logo
    GEMMA-3-27B-IT
    38-GEMMASCOPE-2-TRANSCODER-262K
    INDEX 104
    self-reference
    gpt-5-mini
    Government; literally the system talking to itself!↵↵While our
    Neuronpedia logo
    LLAMA3.1-8B-IT
    19-RESID-POST-AA
    INDEX 110626
    goals
    gpt-5-mini
    applicant about their goals and motivations. This will help you
    Neuronpedia logo
    LLAMA3.1-8B-IT
    19-RESID-POST-AA
    INDEX 57191
    verbs about change
    gpt-5-mini
    floating by and decided to embrace it.↵As she watched
    Neuronpedia logo
    LLAMA3.1-8B-IT
    19-RESID-POST-AA
    INDEX 27476
    I
    gpt-5-mini
    to do?↵A: I haven't decided yet.
    Neuronpedia logo
    LLAMA3.1-8B-IT
    15-RESID-POST-AA
    INDEX 27631
    garbled text
    gpt-5-mini
    13. 【背��将进酒全文
    Neuronpedia logo
    LLAMA3.1-8B-IT
    15-RESID-POST-AA
    INDEX 112241
    numbers
    gpt-5-mini
    .3↵↵%↵↵$↵↵336,206↵↵$↵↵266
    Neuronpedia logo
    LLAMA3.1-8B-IT
    15-RESID-POST-AA
    INDEX 40599
    No
    gpt-5-mini
    shook her head. "No, I appreciate the offer
    Neuronpedia logo
    LLAMA3.1-8B-IT
    15-RESID-POST-AA
    INDEX 99975
    Cyrillic а
    gpt-5-mini
    RuntimeException?↵↵A:↵↵Да. Это легко проверить
    Neuronpedia logo
    LLAMA3.1-8B-IT
    15-RESID-POST-AA
    INDEX 34111
    rogue
    gemini-2.5-flash-lite
    American Eagle):** Typically 1 troy ounce of
    Neuronpedia logo
    GEMMA-3-27B-IT
    16-GEMMASCOPE-2-RES-65K
    INDEX 15
    stopwords
    gpt-5-mini
    dipping, and he knew that it was against the rules
    Neuronpedia logo
    LLAMA3.1-8B-IT
    15-RESID-POST-AA
    INDEX 124902
    and
    gpt-5-mini
    walked to his car, and after a minute, he
    Neuronpedia logo
    LLAMA3.1-8B-IT
    15-RESID-POST-AA
    INDEX 63494
    place names
    gpt-5-mini
    What can I do in NYC iF I only have
    Neuronpedia logo
    LLAMA3.1-8B-IT
    15-RESID-POST-AA
    INDEX 129304
    personal names
    gpt-5-mini
    . Get enough sleep: NAME_1 for 7
    Neuronpedia logo
    LLAMA3.1-8B-IT
    19-RESID-POST-AA
    INDEX 53974
    wants and requests
    gpt-5-mini
    Yes, it is possible to go to the city from
    Neuronpedia logo
    LLAMA3.1-8B-IT
    19-RESID-POST-AA
    INDEX 71061
    function words
    gpt-5-mini
    ABA) to use the same routing number format, which
    Neuronpedia logo
    LLAMA3.1-8B-IT
    19-RESID-POST-AA
    INDEX 73395
    needs and concerns
    gpt-5-mini
    life. It is our desire to be effective and fair
    Neuronpedia logo
    LLAMA3.1-8B-IT
    19-RESID-POST-AA
    INDEX 87531
    hedging language Method used: 2 — detects hedging language
    gpt-5-mini
    that are not likely to be well-known in the target
    Neuronpedia logo
    LLAMA3.1-8B-IT
    19-RESID-POST-AA
    INDEX 111683
    positive sentiment Method used: 2
    gpt-5-mini
    Technical breakout↵10. Economic recovery↵↵Events before a
    Neuronpedia logo
    LLAMA3.1-8B-IT
    19-RESID-POST-AA
    INDEX 59638