© 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-3-270M
    3. 12-GEMMASCOPE-2-RES-16K
    4. 4141
    Prev
    Next
    INDEX
    Explanations

    thoughtsThe user wants a concise explanation (3-20 words) for a neuron's behavior, based on the provided lists.The explanation should identify a pattern related to what the neuron detects or predicts.Let's break down the inputs:**<MAX_ACTIVATING_TOKENS>**:- `I` (3 times)- `you` (5 times)- `she` (1 time)This list heavily features personal pronouns, specifically first and second person.**<TOKENS_AFTER_MAX_ACTIVATING_TOKEN>**:- `was` (after 'I')- `step` (after 'you')- `go` (after 'you')- `think` (after 'you')- `lived` (after 'she')- `'` (punctuation, after 'you')- `first` (after 'I')- `see` (after 'I')These tokens suggest states of being ("was", "lived"), actions ("step", "go"), thoughts ("think"), or initial experiences ("first", "see"). This aligns with personal experiences or actions directed at someone.**<TOP_POSITIVE_LOGITS>**:- `released`- `launch`- `crash`- `arni` (misspelling of Arnie? or possibly some technical term)- `asked`- `released`- `questioned`- `ruz` (likely part of a name or jargon)- `comes`- `queried`This list points towards events, actions, questions, or inquiries.**<TOP_ACTIVATING_TEXTS>**:- "I wish I had this when I was learning." (Past personal experience, learning)- "when you step through the doors, you're transported..." (Instruction/description for 'you', sensory experience)- "before you go and pack accordingly." (Instruction for 'you')- "what's the first thing that comes to mind when you think of a person?" (Question to 'you')- "she used to have allergies when she lived in Seattle but she thinks they are worse here." (Past personal experience of 'she', comparison)- "after you've finished playing it, then you'll find it to be a truly exceptional work of art." (Recommendation/description for 'you')- "I must admit that I have been drawn to you since I first saw you." (Past personal experience/feeling of 'I' towards 'you')- "whenever I see anything related to S, I feel negative emotions because I associate it with the controversy..." (Past personal experience/feeling of 'I')**Pattern Identification**:1. **Pronouns**: The most prominent feature is the heavy use of "I" and "you" in `MAX_ACTIVATING_TOKENS` and `TOP_ACTIVATING_TEXTS`. This neuron is strongly activated by contexts involving the speaker's perspective or direct address to the listener.2. **Personal Experience/Action**: The tokens *after* the pronouns and the content of the texts suggest descriptions of past states ("was", "lived"), personal thoughts/feelings ("think"), initial perceptions ("first", "see"), or actions/instructions directed at "you" ("step", "go").3. **Events/Questions**: The `TOP_POSITIVE_LOGITS` suggest that these personal narratives might lead into or be related to events, inquiries, or questions. past personal experiences and thoughts

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-270m-pt/resid_post/layer_12_width_16k_l0_medium
    Prompts (Dashboard)
    392,802 prompts, 256 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    抜群
    1.27
    preserved
    1.27
    datad
    1.25
    nosed
    1.20
    therefore
    1.15
    worth
    1.08
    since
    1.06
    privileged
    1.06
    あとは
    1.05
     abund
    1.02
    POSITIVE LOGITS
    ENTA
    2.05
     launch
    2.04
     发布
    1.97
     ไหร่
    1.94
    ئي
    1.94
     arrives
    1.92
    ئ
    1.90
     enters
    1.89
     release
    1.88
     Inaug
    1.86
    Activations Density 0.303%

    No Known Activations