© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it
    Neuronpedia
    Jacobian LensNEW
    Natural Language
    Autoencoders
    MAY 26
    Assistant AxisJAN 26Circuit TracerAPR 26SteerSAE EvalsExportsinterp-engineNEW Community BlogPrivacy & TermsContact
    1. Home
    2. Gemma-3-27B-IT
    3. 27-GEMMASCOPE-2-TRANSCODER-262K
    4. 121579
    Prev
    Next
    INDEX
    Explanations

    A list of texts that strongly activate the neuron. - Several texts mention "Morning" or "mornings" in contexts like "Morning is good for setting the tone for the day", "Morning, before bed, or during a lunch break are popular choices", "Morning (Before Work/Studies)", "once in the morning, once in the evening", "early morning or late evening", "likely mornings for an INTJ", "Are you a morning person or a night owl?". - The Spanish text mentions "Mañana" which means "morning". - The theme of "morning" is very strong and consistent across these examples.**Pattern Identification:**The `MAX_ACTIVATING_TOKENS` predominantly consist of variations of the word "morning" (English and Spanish).The `TOP_ACTIVATING_TEXTS` also heavily feature the word "morning" in various contexts, often discussing time management, routines, or daily schedules.The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` sometimes show continuations like "is", "or", "for", suggesting how "morning" is used in phrases.The `TOP_POSITIVE_LOGITS` seem unrelated or very abstract.The strongest and most consistent pattern across the key activating lists (`MAX_ACTIVATING_TOKENS` and `TOP_ACTIVATING_TEXTS`) is the concept of "morning".**Explanation Formulation:**Based on the overwhelming evidence for "morning" in both the activating tokens and texts, the neuron appears to be related to this concept.Constraint Checklist & Confidence Score:1. Concise explanation (3-20 words): Yes.2. Captures what neuron detects/predicts: Yes, by finding patterns.3. Based on finding patterns: Yes.4. In a single short phrase: Yes.5. About tokens following/preceding: Considered, but the core pattern is the token itself.6. About words starting with sequence: Not directly applicable here.7. Avoid listing all tokens: Yes.8.morning routines

    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_27_width_262k_l0_small_affine
    Prompts (Dashboard)
    238,145 prompts, 512 tokens each
    Dataset (Dashboard)
    lmsys + oasst1
    No Configuration Found
    Embeds
    IFrame
    Link
    Negative Logits
     eccess
    0.41
     tjen
    0.38
    靼
    0.38
     kelt
    0.37
    빽
    0.37
    umni
    0.37
    ଭ
    0.37
     éxitos
    0.36
     år
    0.36
     predictor
    0.36
    POSITIVE LOGITS
          
    0.38
    ban
    0.37
    gom
    0.37
    まずは
    0.37
    CMake
    0.36
    #}
    0.36
     Digestive
    0.36
     Sanitary
    0.35
    ixe
    0.35
    brushes
    0.34
    Activations Density 0.003%

    No Known Activations