© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    APIAssistant AxisNEWCircuit TracerNEWSteerSAE EvalsExports Community BlogPrivacy & TermsContact
    1. Home
    2. Gemma-3-27B-IT
    3. 38-GEMMASCOPE-2-TRANSCODER-262K
    4. 390
    Prev
    Next
    INDEX
    Explanations

    The user wants a concise explanation (3-20 words) for a neuron's behavior. This explanation should be a phrase, not a full sentence, and should capture what the neuron detects or predicts by finding patterns. I should avoid phrases like "This neuron detects/predicts", "tokens", "patterns", and capitalization unless it's a proper noun.I have the following inputs:- `MAX_ACTIVATING_TOKENS`: " the ", " , ", " ’ "- `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: "* This\n<end_of_turn>\nfollowing\n1 model\nand\nmodel s"- `TOP_POSITIVE_LOGITS`: Bertha, yuk, тті, общественных, इसरो, obsol, lệnh, unstructured, іх, ക്കേ- `TOP_ACTIVATING_TEXTS`: Examples related to clickbait, story titles, headlines, and some general text.Let's try to find a pattern.1. **Analyze `MAX_ACTIVATING_TOKENS` and `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`**: - `MAX_ACTIVATING_TOKENS`: " the ", " , ", " ’ " - These are common words/punctuation. - `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: "* This", "following", "1 model", "and", "model s" This combination doesn't immediately reveal a strong linguistic pattern in terms of semantic meaning, but it shows punctuation/common words followed by specific content. It might be related to sentence structure or parsing.2.clickbait headlines

    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_38_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
     muziek
    0.45
     music
    0.43
    音樂
    0.41
     музи
    0.40
    music
    0.40
     музы
    0.40
    音乐
    0.40
    but
    0.39
    ceptor
    0.39
    編曲
    0.38
    POSITIVE LOGITS
     Bertha
    0.40
     yuk
    0.40
    тті
    0.37
     общественных
    0.37
     इसरो
    0.36
     obsol
    0.36
     lệnh
    0.35
     unstructured
    0.35
     іх
    0.34
    ക്കേ
    0.34
    Activations Density 0.001%

    No Known Activations