© 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. 37-GEMMASCOPE-2-TRANSCODER-262K
    4. 202062
    Prev
    Next
    INDEX
    Explanations

    **Continuing Analysis of Neuron Behavior**I am analyzing the provided lists to identify overarching themes and recurring elements. My process involves looking for commonalities in sequences, word types, and associated concepts to formulate a precise, word-limited explanation for the neuron's detected function. The subsequent step will be to generate the concise phrase encapsulating this behavior.`TOP_POSITIVE_LOGITS`: `severe`, `severity`, `嚴重`, `prolong` (implied from context, `Prolong` is in `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`) * The top positive logits strongly suggest concepts related to severity, seriousness, and intensity. `嚴重` is Chinese for "severe" or "serious".4. **TOP_ACTIVATING_TEXTS**: * "...if chronic, can contribute to weight loss." * "Potential Tissue Damage: Prolonged restriction can lead to more serious damage." * "...if severe!" * "BOAS significantly reduces their quality of life and lifespan."**Finalizing Neuron Behavior Explanation**I have reviewed the `TOP_POSITIVE_LOGITS`, `TOP_ACTIVATING_TEXTS`, `MAX_ACTIVATING_TOKENS`, and `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` lists. The recurrence of terms like "severe," "severity," and related concepts ("serious damage," "chronic," "prolonged") in both the predicted outputs and example activations strongly indicates the neuron is sensitive to states of intensity, seriousness, or extremity.Therefore, the explanation for this neuron's behavior is:**severity of conditions**

    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_37_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
    Much
    0.39
    dding
    0.37
     poche
    0.37
     многое
    0.36
     Poche
    0.36
     much
    0.36
    ದ್ಧ
    0.36
     Much
    0.35
    ську
    0.35
     pomocą
    0.34
    POSITIVE LOGITS
     severe
    1.08
    severe
    1.03
    Severe
    1.03
     Severe
    0.95
     severity
    0.82
    嚴重
    0.81
    severity
    0.74
     Severity
    0.74
    严重的
    0.73
    Severity
    0.70
    Activations Density 0.016%

    No Known Activations