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

    Neuronpedia

    Natural Language
    Autoencoders
    NEW
    Assistant AxisNEWCircuit TracerUPDATESteerSAE EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Google DeepMind · Exploring Gemma 2 with Gemma Scope
    3. Gemma-2-9B
    4. Residual Stream - 16k
    5. 20-GEMMASCOPE-RES-16K
    6. 13516
    Prev
    Next
    INDEX
    Explanations

    contrastive conjunctions that indicate a divergence or comparison in ideas

    oai_token-act-pair · gpt-4o-miniTriggered by @bot

    two-part correlative constructions that present an initial clause and then emphasize or expand it with a contrasting/additive second clause (e.g., the classic “not X … but Y” pattern).

    oai_token-act-pair · gpt-5Triggered by @jyhe0408

    The neuron activates strongly for the word "but" when it introduces a contrasting or additional element in a sentence, particularly after a comma, and especially when followed by words like "also," "as," "into," "to," or other continuation phrases.

    oai_token-act-pair · claude-4-5-sonnetTriggered by @jyhe0408
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GEMMA-2-9B @ 20-gemmascope-res-16k
    Configuration
    google/gemma-scope-9b-pt-res/layer_20/width_16k/average_l0_68
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    16,384
    Data Type
    float32
    Hook Name
    blocks.20.hook_resid_post
    Hook Layer
    20
    Architecture
    jumprelu
    Context Size
    1,024
    Dataset
    monology/pile-uncopyrighted
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     InputDecoration
    -0.60
     awakeFromNib
    -0.58
    مصادر
    -0.57
    getMinutes
    -0.55
     للمعارف
    -0.54
    edom
    -0.54
    togen
    -0.54
     Kitty
    -0.54
     propOrder
    -0.53
     Profile
    -0.52
    POSITIVE LOGITS
     also
    0.57
     lisäksi
    0.50
     także
    0.49
    also
    0.48
     myös
    0.41
     nejen
    0.41
     dessutom
    0.40
     dokonce
    0.40
     nemlig
    0.39
     además
    0.39
    Activations Density 0.070%

    No Known Activations