© 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-3-27B-IT
    3. 3-GEMMASCOPE-2-TRANSCODER-262K
    4. 88212
    Prev
    Next
    INDEX
    Explanations

    - Mentions "Mint" (a budgeting app).- Mentions code snippets with curly braces and semicolons.- Mentions "Airport" in context of international travel.- Mentions "Arrival/Departure".Let's re-evaluate based on the MOST RECENT prompt instructions."Your response should be a concise explanation (3 to 20 words) that captures what the neuron detects or predicts by finding patterns in lists.""Look at the lists and find a pattern that helps you explain the behavior of the neuron.""The explanation could be about tokens following or preceding certain tokens.""The explanation could be about words starting with a sequence.""Avoid simply listing all the tokens. Instead, try to find patterns."**MAX_ACTIVATING_TOKENS:**- uit- out- uit- International- out- out- out- International- International- out**TOKENS_AFTER_MAX_ACTIVATING_TOKEN:**- .- .- Airport- ;- ;- ;- Airport- Airport- ;Let's combine the observations:1. **'uit' / 'out' followed by '.'**: This appears in contexts like code formatting (`"});`, `.";`, `";`). The `uit` and `out` might be part of larger words or tokens that end up being interpreted as parts of code or structured text.2. **'International' followed by 'Airport'**: This is evident from "Incheon International Airport".The `TOP_POSITIVE_LOGITS` list is quite noisy, containing non-English characters (`ج`, `០០`), Spanish (`aría`, `corresponde`), French (`construire`), punctuation (`.";`, `";`, `"});`), and `fte` which could be part of code. This suggests the neuron might be sensitive to structured text, potentially code or formal documents, where specific phrases like "International Airport" appear alongside code-like markers

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-27b-it/transcoder_all/layer_3_width_262k_l0_small_affine
    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
    K
    0.94
    Y
    0.83
    कर
    0.77
    ↵↵
    0.75
    W
    0.74
    aging
    0.72
    ものは
    0.71
    ่ง
    0.70
    ٹھ
    0.69
    Ind
    0.66
    POSITIVE LOGITS
     construire
    0.68
    .";
    0.67
    ";
    0.63
    fte
    0.63
    ج
    0.63
    ?
    0.62
    aría
    0.60
     corresponde
    0.60
    "});
    0.60
    ០០
    0.59
    Activations Density 0.001%

    No Known Activations