© 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. 13802
    Prev
    Next
    INDEX
    Explanations

    phrases containing past tense forms of "have" indicating experiences or actions

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

    modal perfect constructions expressing counterfactual or hypothetical past actions (… have + past participle, often with a modal).

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

    The neuron activates on the phrase "would have" followed by a past participle, detecting counterfactual or hypothetical statements about what could have happened in the past.

    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
    guiente
    -0.58
     ainfi
    -0.57
     meur
    -0.55
    LEGGI
    -0.55
     voici
    -0.54
     chrétien
    -0.54
    cías
    -0.52
     hiér
    -0.52
     seuil
    -0.52
     règne
    -0.51
    POSITIVE LOGITS
    hloromethane
    0.57
    llac
    0.57
     transported
    0.56
     adsorbed
    0.55
    oudre
    0.54
    burnt
    0.54
    imread
    0.54
    chassis
    0.53
    macher
    0.52
     transporting
    0.52
    Activations Density 0.098%

    No Known Activations