© 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
    3. Gemma-2-2B
    5. 21-GEMMASCOPE-TRANSCODER-16K
    7. 3635
    Prev
    Next
    INDEX
    Explanations

    This neuron seems to be firing on a wide variety of things, but most strongly on mentions of people's names, especially when coupled with titles or relationships

    oai_token-act-pair · gemini-2.0-flash

    past tense verbs

    np_token-act-pair-logits · gpt-4o-mini

    past tense verbs

    np_max-act-logits · gemini-2.0-flash

    say "verbs" Method used: 2 — because tokens after the max token are predominantly action words

    np_max-act-logits · gpt-5-miniTriggered by @chenshw0109
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2b-pt-transcoders/layer_21/width_16k/average_l0_13
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    16,384
    Data Type
    float32
    Hook Name
    blocks.21.ln2.hook_normalized
    Architecture
    jumprelu_transcoder
    Context Size
    1,024
    Dataset
    monology/pile-uncopyrighted
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    tvguidetime
    -0.95
     purpoſe
    -0.79
     Efq
    -0.78
     Monfieur
    -0.77
    ScopeManager
    -0.75
     Jefus
    -0.74
     myſelf
    -0.70
    SequentialGroup
    -0.69
     uſe
    -0.68
     ſtate
    -0.67
    POSITIVE LOGITS
     acted
    1.06
     kept
    1.02
     walked
    1.00
     showed
    0.98
     ran
    0.98
     looked
    0.97
     responded
    0.96
     tended
    0.96
     appeared
    0.96
     went
    0.95
    Activations Density 6.939%

    No Known Activations