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

    Neuronpedia

    APIAssistant AxisNEWCircuit TracerNEWSteerSAE EvalsExports Community BlogPrivacy & TermsContact
    1. Home
    2. Qwen3-1.7B
    3. 27-LLAMASCOPE-2-LORSA-16K-K64
    4. 16011
    Prev
    Next
    INDEX
    Explanations

    <THINKING> Method 1: MAX_ACTIVATING_TOKENS contain many references to fibrous minerals and asbestos (e.g., “fibrous varieties”, “asbestos”, “amphibole group minerals”). This suggests the neuron is tuned to asbestos terminology. Method 2: Tokens after max-activating tokens are varied, no clear pattern. Method 3: TOP_POSITIVE_LOGITS are dominated by “asbestos”, “fibers”, and related terms, confirming the asbestos focus. Methods 4 and 5 do not add new insight. The clearest pattern comes from Method 3 (reinforced by Method 1): the neuron strongly predicts asbestos‑related words. Explanation: asbestos </THINKING> asbestos

    unknown · unknown
    New Auto-Interp
    Top Features by Cosine Similarity
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    intros
    -18.25
    侑
    -17.25
     JDBC
    -17.25
     Klopp
    -17.13
    Bias
    -16.75
     élect
    -16.63
    逵
    -16.50
    SqlServer
    -16.13
    _bias
    -16.13
     bias
    -16.13
    POSITIVE LOGITS
     asbestos
    21.75
    bestos
    17.75
     fibers
    16.63
    纤维
    15.69
     fiber
    15.44
    丝绸
    15.06
    fib
    14.94
    chine
    14.88
    美
    14.63
    缂
    14.44
    Activations Density 0.029%

    No Known Activations