© 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 EvalsExportsinterp-engineNEWAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Gemma-3-27B-IT
    3. 2-GEMMASCOPE-2-TRANSCODER-262K
    4. 107978
    Prev
    Next
    INDEX
    Explanations

    example, there are 4 rules in total, 10 points for each rule, and then calculate the average score""For example, when her neck is touched, or something like that. How can I do that?""For example, are you: * **Looking for help with a specific coding problem?**""For example, given an image of a man struggling to stay afloat in water, we can reason that the man fell into the water sometime""For example, you could say: * "What is the time in London?"""For example, do you want it to be: * **More formal?**""For example, the code `3` puts the number 3 on the stack. What do you think the code `3 4""For example, it would be helpful to know: * **What is the issue?**""For example, is it optimal to have a single LLM model that understands every single language or to have multiple smaller models that each specialize""For example, typing **bold** will instantly make the text bold."The phrase "For example," appears very frequently in the `TOP_ACTIVATING_TEXTS`.The `MAX_ACTIVATING_TOKENS` are all commas. Commas are often used before or within phrases like "for example".The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` list includes words that can follow a comma or a pause.The `TOP_POSITIVE_LOGITS` list also includes punctuation like `’` and `”`, which are often associated with sentence structure and punctuation.The most prominent pattern is the use of "for example," to introduce illustrative cases or explanations. The neuron seems to be picking up on this common introductory phrase.Therefore, the neuron is likely detecting or predicting the phrase "for example".It needs to be 3-20 words."for example" is 2 words.This fits the criteria.for example

    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_2_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
    𝟎
    1.46
    ه
    1.33
    া
    1.32
    ла
    1.30
    0
    1.30
    ாக
    1.26
    0
    1.20
    기
    1.18
    ме
    1.16
    ك
    1.16
    POSITIVE LOGITS
    ’
    1.31
    And
    0.98
    Z
    0.97
    K
    0.93
    F
    0.92
    H
    0.92
    B
    0.89
    ung
    0.89
    U
    0.87
    ”
    0.85
    Activations Density 0.018%

    No Known Activations