© 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
    3. 31-GEMMASCOPE-2-RES-262K
    4. 26488
    Prev
    Next
    INDEX
    Explanations

    scientific notation and units like K, s, or numbers separated by commas.2. p-values (e.g., p<0.05).3. quantities (e.g., 0 μg/ml, 80-95%).The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` list confirms this: `0`, `K`, `s`, `~`.The `TOP_POSITIVE_LOGITS` list might suggest things like "il" or "er" which could be part of scientific terms or abbreviations.Given the strong presence of numbers, scientific units (K, s), and p-values, the neuron seems to be related to reporting numerical or statistical findings. The `<0x89>` token often acts as a separator before these values or units.A concise explanation for this would be "numerical values and scientific units".This fits the word count and the pattern observed.Let's refine: it's not just *any* numerical value, but often statistical or quantitative measures.Consider "statistical values and units".Or "quantities and units".Looking at the TOP_POSITIVE_LOGITS, 'il' could be 'mil' or 'µl'. 'K' is clearly Kelvin. 's' is seconds. '0' is zero.The pattern is reporting quantities, scientific values, and their associated units or statistical significance."quantities and scientific notation""numerical and statistical reporting""values with units or significance"Let's go with "numerical values and scientific units" as it covers the most direct observations. numerical values and scientific units

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

    mathbb

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-27b-pt/resid_post/layer_31_width_262k_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
    ي
    0.90
    <0x0D>
    0.79
    𝘢
    0.76
    리
    0.75
    </h2>
    0.74
    ні
    0.73
    ﻲ
    0.72
    و
    0.71
    কে
    0.70
    يها
    0.69
    POSITIVE LOGITS
    il
    0.92
    er
    0.75
    <0x9F>
    0.73
    <0xBC>
    0.72
    ag
    0.69
    <0xAB>
    0.68
    ar
    0.65
    <0xBD>
    0.64
    <0xAE>
    0.64
    <0xB9>
    0.63
    Activations Density 0.000%

    No Known Activations