© 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-262K
    4. 2694
    Prev
    Next
    INDEX
    Explanations

    The explanation must be 3 to 20 words.The explanation should be a single phrase, not a full sentence.It should capture what the neuron detects or predicts by finding patterns.Looking at:- `TOP_POSITIVE_LOGITS`: `, same`, `(“`, `,【`, `ری`, `ਾ`, `}`, `maßen`, `ло`, `ྔ`- `TOP_ACTIVATING_TEXTS`: Contains snippets of code (JavaScript, Python, VBA, Java), HTML, and text descriptions. These include examples of code blocks ending, function returns, handling errors, and descriptive text.The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` (return, }, };) strongly suggests code constructs.The presence of foreign characters and symbols in `TOP_POSITIVE_LOGITS` is a bit noisy but doesn't negate the code aspect seen in the activating texts.The "return" and closing brace `}` appearing after the max activating token, and the examples of code blocks in the activating texts, point towards the neuron activating for code-related segments, particularly around function endings or returns.Let's re-evaluate based on the prompt rules:- Explain behavior of a neuron.code structure markers

    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_262k_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
     and
    1.58
     or
    1.51
     
    1.50
     to
    1.48
     which
    1.45
     that
    1.23
     on
    1.19
     a
    1.17
     their
    1.15
     the
    1.12
    POSITIVE LOGITS
    ,
    1.09
    zelfde
    1.08
    (“
    1.00
    ,【
    0.98
    ری
    0.96
    ਾ
    0.96
    TorpedoStore
    0.95
    maßen
    0.91
    ло
    0.90
    ྔ
    0.90
    Activations Density 1.418%

    No Known Activations