© 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
    1. Home
    2. Gemma-3-27B-IT
    3. 31-GEMMASCOPE-2-RES-262K
    4. 128336
    Prev
    Next
    INDEX
    Explanations

    listing specific examples*Self-correction during thought process:*The words `particularly` and `especially` appear in `MAX_ACTIVATING_TOKENS`.The subsequent tokens are often prepositions or phrases indicating description or scope, like `in`, `to`, `with`, `or`, `customer`, `financial`, `parameters`, `color`.The `TOP_ACTIVATING_TEXTS` frequently contain phrases like "particularly in areas related to...", "list types of information - e.g., business plans, customer lists, financial information...", "Trying to self-treat...", "Same parameters as first call...", "especially with Dyonasian aspect...".The core pattern seems to be introducing a specific detail or list of items/examples after a modifying word or a verb indicating effort/action.Let's refine:- `particularly` + `in areas related to X, Y, Z.`- `Trying` + `to do something.`- `[list types of information - e.g., A, B, C]`- `sleeves` (context for item description)- `Same parameters`This neuron seems to be about specifying details or providing particular examples.The `TOP_POSITIVE_LOGITS` (`Hilo`, `Sac`, `Aqu`, `Humidity`, `Staff`, `Colombo`, `Arquivo`, `ACO`, `Salon`, `rosters`) are not strongly related to this pattern of listing specifics. They are more like names or labels.Considering `MAX_ACTIVATING_TOKENS`: `particularly`, `Trying`, `Same`, `sleeves`, `especially`.Considering `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: `in`, `customer`, `to`, `parameters`, `financial`, `color`, `or`, `with`.Considering `TOP_ACTIVATING_TEXTS`: many examples involve listing specific types of information, physical attributes, or actions. listing specific examples

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-27b-it/resid_post/layer_31_width_262k_l0_medium
    Prompts (Dashboard)
    238,145 prompts, 512 tokens each
    Dataset (Dashboard)
    lmsys + oasst1
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    ซึ่ง
    0.45
     März
    0.44
    arın
    0.43
     "$
    0.42
     który
    0.41
    ศาส
    0.41
    __["
    0.41
     যৌ
    0.41
    (".
    0.40
     yılında
    0.40
    POSITIVE LOGITS
    Hilo
    0.50
    Sac
    0.49
    Aqu
    0.48
    Humidity
    0.47
    Staff
    0.46
    Colombo
    0.45
    Arquivo
    0.45
    ACO
    0.45
    Salon
    0.45
     rosters
    0.44
    Activations Density 0.003%

    No Known Activations