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

    Neuronpedia

    APIAssistant AxisNEWCircuit TracerNEWSteerSAE EvalsExports Community BlogPrivacy & TermsContact
    © Neuronpedia 2025
    Privacy & TermsBlogGitHubSlackTwitterContact
    1. Home
    2. Gemma-3-12B
    3. 12-GEMMASCOPE-2-RES-16K
    4. 0
    Prev
    Next
    INDEX
    Explanations

    sustainability electives vehicles

    np_acts-logits-general · gemini-2.5-flash-lite

    The main thing this neuron does is find single letters serving as initials or the first character of capitalized words, acronyms, or specific identifiers.

    oai_token-act-pair · gemini-2.5-flashTriggered by @hijohnnylinnewwww

    This neuron appears to be malfunctioning or capturing noise rather than finding a meaningful linguistic pattern. The activations are sporadic, scattered across completely unrelated tokens (proper nouns, fragments, punctuation) in diverse text genres (gambling sites, legal documents, coffee shops, sports articles

    oai_token-act-pair · claude-4-5-haikuTriggered by @hijohnnylinnewwww

    capitalized letters that begin proper nouns or brand names.

    oai_token-act-pair · claude-4-5-sonnetTriggered by @hijohnnylinnewwww
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-12b-pt/resid_post/layer_12_width_16k_l0_medium
    Prompts (Dashboard)
    392,802 prompts, 256 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    ש
    1.62
    et
    1.54
    and
    1.41
    ن
    1.39
    en
    1.36
    el
    1.27
    ال
    1.26
    ла
    1.21
    at
    1.20
    то
    1.20
    POSITIVE LOGITS
    mobilpay
    0.99
     ومع
    0.97
    ৯
    0.86
     ouvert
    0.84
    fungsi
    0.84
    ৭
    0.83
     invas
    0.83
    を有
    0.82
    ுள்ளனர்
    0.82
    dokter
    0.82
    Activations Density 0.819%

    No Known Activations