© 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-4-31B
    3. 30-RES-MATRYOSHKA-131K
    4. 118197
    Prev
    Next
    INDEX
    Explanations

    This neuron represents the concept of similarity or comparison, highlighting relationships and likenesses between different entities or ideas. It frequently appears in contexts where one thing is being likened to another, whether structurally, conceptually, or in terms of function.Let's break down the provided lists to find the pattern:* **TOP_POSITIVE_LOGITS**: This list is dominated by synonyms for "similar" and "comparable" in various languages (English, Spanish, Russian, Japanese). This strongly suggests the neuron is related to the concept of similarity or comparison.* **TOP_ACTIVATING_TEXTS**: * "...compare our THFA DCS" - Explicitly uses "compare". * "...Is this situation similar to..." - Explicitly uses "similar". * "...structurally distinct from the" - Implies comparison based on structural properties, a form of similarity/difference detection. * "...between this and Mitch McConnell’s allies..." - Implies comparison or contrast. * "Governors Island as a pair of conjoined pancakes..." - Uses "as" to draw a comparison.* **MAX_ACTIVATING_TOKENS**: "FA", "yet", "and", "similar", "this", "as", "structurally", "this", "their", "it". The presence of "similar" and "structurally" aligns with the positive logits and text examples. "as" is also a common word used for comparison.* **TOKENS_AFTER_MAX_ACTIVATING_TOKEN**: * FA **DCS** (no clear pattern from this one) * yet **I** (no clear pattern) * and **the** (general) * similar **to** (very common pairing for similarity) * this **relate** (hints at relationship/comparison) * as **a** (general) * structurally **distinct** (pairing that emphasizes difference, hence comparison) * this **and** (general) * their **the** (general) * it **the** (general)The strongest pattern is clearly the concept of similarity, comparison, or detecting likeness and difference.**Explanation:** Detecting similarity and comparison

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    decoderesearch/gemma-4-saes/gemma-4-31b
    Prompts (Dashboard)
    16,384 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     regole
    -0.07
     permesso
    -0.06
     inquiring
    -0.06
     these
    -0.06
     ambiguities
    -0.05
     inquisitive
    -0.05
    เหล่านี้
    -0.05
    finders
    -0.05
     সমাজের
    -0.05
    eil
    -0.05
    POSITIVE LOGITS
     similar
    0.12
     comparable
    0.11
    similar
    0.11
    Similar
    0.11
     analogous
    0.10
     similares
    0.10
     аналоги
    0.10
     Similar
    0.10
     podob
    0.10
    と同様
    0.09
    Activations Density 0.055%

    No Known Activations