© 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. 15-GEMMASCOPE-2-TRANSCODER-262K
    4. 15112
    Prev
    Next
    INDEX
    Explanations

    The neuron detects common English pronouns followed by verbs, indicating actions or states.**Explanation:** A pronoun followed by a verb.Let's re-evaluate based on the rules and examples. The rule is to *find a pattern*.MAX_ACTIVATING_TOKENS: She, I, He, we, ., She, ., Can, It, SheTOKENS_AFTER_MAX_ACTIVATING_TOKEN: has, had, knows, have, Please, had, Can, you, d, hadLet's look at the pairings:* She has* I had* He knows* we have* She had* She had* Can youThe TOP_POSITIVE_LOGITS are Chinese, Arabic, Russian, Italian, Hindi, Bengali. This is highly discordant with the MAX_ACTIVATING_TOKENS which are all English pronouns. This suggests the neuron might be detecting English pronoun-verb pairings, and the positive logits are noise or perhaps some rare cross-lingual association if the underlying model is multilingual.pronoun 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_15_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
    sk
    0.22
    ljenje
    0.20
     subsection
    0.20
     averse
    0.19
    nsec
    0.19
     editorials
    0.19
     OSError
    0.18
    ösung
    0.18
     DURING
    0.18
     inextricably
    0.18
    POSITIVE LOGITS
    營養
    0.21
    ؟
    0.20
    因为
    0.20
     هذا
    0.19
     исторических
    0.19
     итальян
    0.19
    लीवुड
    0.19
    因为它
    0.19
     इज़
    0.19
    Bài
    0.18
    Activations Density 0.040%

    No Known Activations