© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    Natural Language
    Autoencoders
    NEW
    Assistant AxisNEWCircuit TracerUPDATESteerSAE EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Gemma-3-27B-IT
    3. 31-GEMMASCOPE-2-RES-65K
    4. 8526
    Prev
    Next
    INDEX
    Explanations

    The pattern identifies tokens that introduce or establish a character's role, persona, or identity at the beginning of roleplay scenarios, particularly in user prompts requesting the AI to adopt a specific character. This includes articles, pronouns, and linking words that directly precede or connect to character descriptions (e.g., "a man who," "you are," "the," possessive pronouns). The pattern also captures tokens that signal transitions into character roleplay mode, such as stage directions in parentheses, opening narrative actions, and initial character establishment phrases. Additionally, it marks tokens that emphasize character traits or personality aspects being defined, especially adjectives and descriptors that shape the persona being adopted.

    eleuther_acts_top20 · claude-4-5-sonnetTriggered by @jamesnaruto04

    speaking, telling, or narrating

    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_65k_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
     lacked
    0.45
     решили
    0.38
    できなかった
    0.38
     clever
    0.37
     did
    0.36
     chose
    0.36
     Cheat
    0.36
     decided
    0.36
    得到了
    0.36
     বিত
    0.36
    POSITIVE LOGITS
     erzählt
    0.73
     рассказывает
    0.72
     proudly
    0.61
     рассказа
    0.59
     habla
    0.59
     mówi
    0.59
     parla
    0.59
     интервью
    0.58
     говорит
    0.58
     erzählen
    0.58
    Activations Density 0.201%

    No Known Activations