© 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 EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Gemma-3-27B-IT
    3. 31-GEMMASCOPE-2-TRANSCODER-262K
    4. 216990
    Prev
    Next
    INDEX
    Explanations

    This neuron is related to understanding who is doing what, often followed by verbs.Let's refine this based on the rules.The pattern is "who" followed by a verb that describes an action or state.Examples: "who is", "who enjoys", "who are", "who doesn't".The TOP_POSITIVE_LOGITS include "enjoys", "genießen" (German for enjoy). This reinforces the idea of actions/states.The core pattern seems to be identifying a subject ("who") and then its predicate/action."who is" / "who are""who enjoys" / "who genießen"This sounds like it's about characterizing or describing a person, or asking about their actions/states.Let's consider the explanation:- "who questions" - too short, doesn't capture the *following* part.- "who followed by actions" - uses "followed by", maybe too close to patterns.- "who is or are" - too specific, misses "enjoys".who followed by verb

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-27b-it/transcoder_all/layer_31_width_262k_l0_small_affine
    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
     разработки
    0.39
    肐
    0.39
    работки
    0.37
     frock
    0.37
    ယောက်
    0.37
     Ген
    0.37
     отправи
    0.36
     परवरिश
    0.36
    nour
    0.36
     широко
    0.36
    POSITIVE LOGITS
    当
    0.38
    ☐
    0.38
    眺
    0.35
     enjoys
    0.35
    工业
    0.35
     آگے
    0.35
    ImageBox
    0.35
     genießen
    0.34
     bedrooms
    0.33
     enjoyment
    0.33
    Activations Density 0.001%

    No Known Activations