© 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-4-31B
    3. 30-RES-MATRYOSHKA-131K
    4. 118383
    Prev
    Next
    INDEX
    Explanations

    RS-1: Pediatric dosimetry of 18 F-FDG whole body PET/CT scans Aruna Kaushik, (Contains `Pediatric`, `dosimetry`, `F-FDG`, `PET/CT`) - `47 - "all in this together" - angie checks in at front desk, goes up to room. jake returns to front desk, after getting story straight` (Contains `-`, `ang`) - `+ 63. Suppose -k + 1 = 3. Which is the second smallest value? (a) f (b) k (c) w` (Contains numbers, mathematical expressions, `+`, `-`)Now let's look for patterns.**MAX_ACTIVATING_TOKENS**: `.`, `in`, `the`, `!`, `Pediatric`, `-`**TOKENS_AFTER_MAX_ACTIVATING_TOKEN**: `July`, `So`, `-`, `MySQL`, `peak`, `dosimetry`, `ang`, `3`Observations:- `Pediatric` is in MAX_ACTIVATING_TOKENS. `dosimetry` is in TOKENS_AFTER_MAX_ACTIVATING_TOKEN. This appears in a text related to PET/CT scans.- `in the` is in MAX_ACTIVATING_TOKENS. `So` is in TOKENS_AFTER_MAX_ACTIVATING_TOKEN. `So,` is in a text.- `.` is in MAX_ACTIVATING_TOKENS. `July` is in TOKENS_AFTER_MAX_ACTIVATING_TOKEN. `July 18, 2008` appears in a text.- `-` is in MAX_ACTIVATING_TOKENS. `ang` is in TOKENS_AFTER_MAX_ACTIVATING_TOKEN. `angie` appears in a text.- `-` is in MAX_ACTIVATING_TOKENS. `MySQL` is in TOKENS_AFTER_MAX_ACTIVATING_TOKEN. `MySQL 5.0.22` appears in a text.- `-` is in MAX_ACTIVATING_TOKENS. `peak` is in TOKENS_AFTER_MAX_ACTIVATING_TOKEN. `peak hours` appears in a text.- `-` is in MAX_ACTIVATING_TOKENS. `3` is in TOKENS_AFTER_MAX_ACTIVATING_TOKEN. Number 3 appears in mathematical questions.- `!` is in MAX_ACTIVATING_TOKENS.The common theme seems to be specific words/tokens that appear, and then followed by a more specific word (often more detailed or a sub-category).- `Pediatric` -> `dosimetry`- `.` -> `July` (date marker)- `-` -> `MySQL` (version marker)- `-` -> `peak` (context marker)- `-` -> `ang` (name marker)- `-` -> `3` (number marker)The examples suggest the neuron is capturing tokens that precede a specific piece of information.numbers and labels

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    decoderesearch/gemma-4-saes/gemma-4-31b
    Prompts (Dashboard)
    16,384 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     ponieważ
    -0.08
     because
    -0.07
    Such
    -0.07
    because
    -0.07
     omdat
    -0.06
     drie
    -0.06
    sequently
    -0.06
     trois
    -0.06
     защото
    -0.06
     ક્લિક
    -0.06
    POSITIVE LOGITS
     sẽ
    0.07
    enough
    0.06
     অনুক
    0.06
     kommer
    0.06
     will
    0.06
     Moment
    0.05
     serán
    0.05
     எண்ணிக்க
    0.05
    人群
    0.05
    小
    0.05
    Activations Density 0.037%

    No Known Activations