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

    Neuronpedia

    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. 3154
    Prev
    Next
    INDEX
    Explanations

    The neuron seems to be focusing on specific grammatical structures or punctuation following certain words.- `that` followed by `her` or `'` or `that`.- `.` followed by `Haven` or `of` or `of`.- `of` followed by `of` or `h` or `.`.The `TOP_POSITIVE_LOGITS` are a mix of characters and short text fragments, not immediately forming a clear semantic pattern.The `TOP_ACTIVATING_TEXTS` show examples of:- Explaining possibilities ("one specific way he could have learned it. However, he could have")- Describing emotional states or coping mechanisms ("struggle to regulate their emotions effectively. They might feel overwhelmed by sadness,", "unfulfilling or overwhelming. * **Simply Haven't Found the Right Person:")- Explaining a scenario ("She might be misinterpreting the rejection. Perhaps she thought there was a stronger connection than there was.")- Conversational phrases ("You know, I'm surprised you haven't gotten a medal! A little something for showing up dressed.")- Explaining interpretation ("We have to *reinterpret* "Mafia" to fit this structure. **Possible Interpretations of")- Explaining consequence ("even if it had unintended negative consequences. You're not accepting blame because your motivation was positive. **4. Disclaiming Responsibility")Looking at `MAX_ACTIVATING_TOKENS` and `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`:`that` -> `her` (

    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
     tributaries
    1.20
     Criticism
    1.20
     tannins
    1.17
     Conclusions
    1.16
     
    1.15
     organs
    1.13
     softener
    1.13
     grievances
    1.13
     wrongs
    1.13
     ascribe
    1.13
    POSITIVE LOGITS
    r
    2.22
    v
    2.16
    l
    1.97
    ر
    1.92
    om
    1.80
    m
    1.79
    it
    1.77
    ről
    1.77
    ار
    1.76
    an
    1.73
    Activations Density 0.001%

    No Known Activations