© 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-3-27B-IT
    3. 37-GEMMASCOPE-2-TRANSCODER-262K
    4. 204677
    Prev
    Next
    INDEX
    Explanations

    `named``Organization``patterns``programs``TOP_POSITIVE_LOGITS`:* `reflections`* `rctx`* `ulaires` (likely part of a French word, e.g., "populaires")* `contextual`* `Vietnam`* `Bootcamp`* `securitycenter`* `doğan` (Turkish)* `ິນ` (Thai)* `मलयालम` (Malayalam)This list is very diverse. It includes concepts (`reflections`, `contextual`), specific entities (`Vietnam`, `Bootcamp`, `securitycenter`, `doğan`), and then non-English words (`ulaires`, `ິນ`, `मलयाalam`). This suggests it might be identifying named entities or specific contexts, but the inclusion of non-English words is a strong signal.`TOP_ACTIVATING_TEXTS`:* "move them closer to those goals. Require search and planning algorithms to find the best action sequence. * **...**" (mentions search and planning algorithms)* "... :ARG1 (p / place :domain (d / this))) :op2 (b / be-located-at :ARG1 (s / service)..." (semantic tree/ontology notation)* "... user generate a semantic tree of a natural language text input based on all the named entities in the input: "I want you to implement a physics-driven FABRIK implementation using ArticulationBodies 1547578", "Name": "Patient Care Team", "SemanticTypes": ["Organization"], "Context": "Healthcare Administration - May ) (not (holding ?x)) (clear ?x)) ) (:action stack :parameters (?x - block ?y - block) :precondition (and (holding ?x) (clear ?..." (mentions semantic tree, named entities, Organization, specific actions/states.The neuron is identifying contexts where explicit named entities or organizations are mentioned, often within structured data or specific domains, and potentially including non-English language identifiers

    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_37_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
     artificial
    0.75
     Artificial
    0.73
    Artificial
    0.71
     expert
    0.70
    artificial
    0.69
     Expert
    0.66
    expert
    0.65
     искус
    0.63
    Expert
    0.61
     artificielle
    0.56
    POSITIVE LOGITS
     reflections
    0.38
    rctx
    0.37
    ulaires
    0.37
     contextual
    0.34
    Vietnam
    0.34
     Bootcamp
    0.34
    securitycenter
    0.34
    doğan
    0.34
    ິນ
    0.34
     मलयालम
    0.33
    Activations Density 0.067%

    No Known Activations