© 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. Gemma-3-12B
    3. 24-GEMMASCOPE-2-RES-16K
    4. 7120
    Prev
    Next
    INDEX
    Explanations

    The neuron specializes in detecting technical/log‐style tokens, particularly error messages and numeric or alphanumeric codes (e.g. “Error,” plugin names, version numbers, dollar amounts).

    oai_token-act-pair · o4-miniTriggered by @jyhe0408

    Looking at the activations, this neuron strongly activates on punctuation marks and special characters that separate or denote sections, particularly: - Em dashes (—) with very high activations (596, 221, etc.) - Parentheses, especially opening parenth

    oai_token-act-pair · claude-4-5-sonnetTriggered by @jyhe0408

    formatting and symbolic markers tied to technical or metadata context, such as punctuation, parentheses, arrows, paths, alphanumeric codes, acronyms, and other structured references.

    oai_token-act-pair · gpt-5Triggered by @jyhe0408
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-12b-pt/resid_post/layer_24_width_16k_l0_medium
    Prompts (Dashboard)
    392,802 prompts, 256 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     underpinned
    0.75
     ensues
    0.73
     Featuring
    0.73
     vemos
    0.71
     albeit
    0.71
    ↵↵↵↵
    0.70
     andre
    0.70
    ↵↵↵
    0.70
     Importantly
    0.70
    ↵↵
    0.69
    POSITIVE LOGITS
    datasets
    0.75
    data
    0.72
    dataset
    0.70
    setosa
    0.68
    america
    0.68
    ቈ
    0.68
    onuclease
    0.67
    hentication
    0.65
    mappings
    0.65
    classifiers
    0.64
    Activations Density 1.962%

    No Known Activations