© 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. 40-GEMMASCOPE-2-TRANSCODER-262K
    4. 28710
    Prev
    Next
    INDEX
    Explanations

    active form of vitamin D..." * "...public transportation, walking, biking, or rideshares." * "...P2Y12 receptor inhibitor..." * "...active form of Vitamin A." * "...Prevention of organ rejection.The neuron appears to be sensitive to concepts and words related to "activity" and "forms" (especially in a biological or medical context, like "active form"). The presence of "active" repeatedly in `TOP_POSITIVE_LOGITS`, and its appearance in `MAX_ACTIVATING_TOKENS` and `TOP_ACTIVATING_TEXTS` are strong indicators. The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` list showing `form` multiple times reinforces this.Therefore, a concise explanation would focus on these core elements.- **active** (from MAX_ACTIVATING_TOKENS and TOP_POSITIVE_LOGITS)- **form** (from TOKENS_AFTER_MAX_ACTIVATING_TOKEN)- context of vitamin D, vitamin A, receptor inhibitors, organ rejection (from TOP_ACTIVATING_TEXTS) suggests biological/medical active *forms*.- `walking`, `transportation` (from MAX_ACTIVATING_TOKENS) also relate to being active or in motion.The most consistent and specific pattern seems to be related to "active form", often in a biological or functional sense.Let's check the word count: "active form" is 2 words, which is within the 3-20 word limit.Consider other options:- "active or movement" (3 words) - captures both aspects.- "biological active form" (3 words) - more specific context.- "active states or forms" (4 words) - generalizes well.Looking at `MAX_ACTIVATING_TOKENS`: `active`, `walking`, `.`, `transportation`.Looking at `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: `receptor`, `7`, `form`, `,`, `1`, `,`, `corticosteroids`, `form`.The `active` token is common. The `form` token is common after `active` in some texts (e.g., "active form").The `TOP_POSITIVE_LOGITS` strongly confirm "active" in many forms and languages.The phrase "active form" seems very representative, especially considering texts like "active form of vitamin D", "active form of Vitamin A". `walking` and `transportation` *could* be seen as forms of activity, but `active form` is more directly supported by the `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`.Let's refine:If the neuron detects patterns related to "active form", then the explanation should be that.Alternative interpretation: The neuron activates for "active" and possibly things that *enable* activity or are *in* an active state."active state or form" - 4 words.The phrase "active form" is highly specific and directly observable from the lists.active form

    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_40_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
     acceptor
    0.64
    Accept
    0.62
    ACCEPT
    0.62
     Accept
    0.60
     acceptors
    0.59
    accept
    0.59
     Accepts
    0.55
     accepts
    0.55
     accepting
    0.53
     accett
    0.53
    POSITIVE LOGITS
     active
    0.93
     Active
    0.93
    Active
    0.84
     актив
    0.78
    active
    0.75
     ACTIVE
    0.75
     aktive
    0.67
     aktif
    0.62
     활
    0.61
     reactive
    0.60
    Activations Density 0.042%

    No Known Activations