© 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
    2. Gemma-3-12B-IT
    3. 12-GEMMASCOPE-2-RES-16K
    4. 14005
    Prev
    Next
    INDEX
    Explanations

    thoughtThe user wants a concise explanation (3-20 words) of a neuron's behavior, based on provided lists.The explanation should capture a pattern found in the lists.Let's analyze the provided lists:- **MAX_ACTIVATING_TOKENS**: `Variable`, `conditions`, `Inference`, `MACHINE`, `seeds`, `Thought`, `intelligence`, `Negeri`, `ought`.- **TOKENS_AFTER_MAX_ACTIVATING_TOKEN**: `=`, `:`, `.`, `(`, `!)`, `>`, `assistant`.- **TOP_POSITIVE_LOGITS**: `casework`, `Q`, `X`, `Па`, `activid`, `Frankly`, `Rydberg`, `addling`, `crocodile`, `downg`.- **TOP_ACTIVATING_TEXTS**: These texts often contain phrases like: * `chain_of_thought = []` * `thought = "..."` * `Independent Variable: ...` * `Dependent Variable: ...` * `monetary conditions.` * `Natural Language Inference (NLI),` * `MACHINE.` * `Seed: (try different seeds!)` * `Thought` * `assistant` * `SMA Negeri 1 Palembang` (Negeri) * `Question: ...` * `THought: ...`**Pattern Identification:**1. **MAX_ACTIVATING_TOKENS** contains keywords that often appear in programming contexts, structured data, or specific output formats. * `Variable`, `conditions`, `Inference`, `seeds`, `intelligence`, `ought`. * `MACHINE` suggests a machine agent or system. * `Negeri` shows up in the activating texts as part of `SMA Negeri 1 Palembang`.2.Variable assignment and structured output

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-12b-it/resid_post/layer_12_width_16k_l0_medium
    Prompts (Dashboard)
    238,145 prompts, 512 tokens each
    Dataset (Dashboard)
    lmsys + oasst1
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    𝗶
    0.85
     mirar
    0.79
    𝗲
    0.78
    ード
    0.77
    präsident
    0.76
     róż
    0.74
    iennes
    0.73
     radius
    0.72
    らは
    0.72
    struktur
    0.71
    POSITIVE LOGITS
     casework
    0.88
    Q
    0.88
    X
    0.86
    Па
    0.84
     activid
    0.80
     Frankly
    0.77
     Rydberg
    0.77
    addling
    0.76
    crocodile
    0.76
     downg
    0.75
    Activations Density 0.000%

    No Known Activations