© 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
    3. Gemma-3-27B-IT
    5. 27-GEMMASCOPE-2-TRANSCODER-262K
    7. 194947
    Prev
    Next
    INDEX
    Explanations

    * **MAX_ACTIVATING_TOKENS**: This list is empty.* **TOKENS_AFTER_MAX_ACTIVATING_TOKEN**: This list is empty.The other lists suggest patterns related to web development, UI elements, and links. `TOP_POSITIVE_LOGITS` includes words like 'put', 'reply', 'replies', 'satisfactory'. 'Hammer' and 'felt' are outliers. 'คิด' is Thai for 'think'. The '𝓰' character is strange.Given the empty `MAX_ACTIVATING_TOKENS` and `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`, I cannot use the primary rule about immediate context.The `TOP_ACTIVATING_TEXTS` are heavily skewed towards web page structure: `<nav>`, `<a>`, `href`, `<h1>`, `<div>`, class names like `flex`, `space-x-6`, `hover:text-tan`, `logo`, `YourBrand`, `Purrfect Cats`, `YourLogo`, `Netflix Logo`.The `TOP_POSITIVE_LOGITS` seem less indicative of this structural pattern. 'put', 'reply', 'replies' could relate to form submissions or interactions.Let's prioritize the `TOP_ACTIVATING_TEXTS` as they provide the most direct evidence of content. The neuron seems to be activating on parts of HTML/JSX structures that define navigation, links, and branding elements within web pages.How to describe this concisely?- `nav links and branding` (4 words)- `web page navigation and branding` (5 words)- `website navigation and content structure` (5 words)- `html structural elements and links` (5 words)- `navigation and link elements` (4 words)The logits are a bit confusing. 'satisfactory', 'reply', 'replies' and 'put' might suggest a secondary function related to feedback or user input *within* these pages, but the overwhelming signal is structural/navigational.Considering the rule "You should look at the lists and find a pattern that helps you explain the behavior of the neuron." and "The explanation should be specific."Focusing on the structural HTML/JSX elements and links:"navigation and links" seems too simple given `TOP_ACTIVATING_TEXTS`."website navigation and branding" seems good."web page navigation structure"Let's try to capture the *elements* that define the website structure."nav, links, and website branding"nav links and website branding

    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_27_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
     Traders
    0.37
     escuelas
    0.36
    醫院
    0.35
     Tourism
    0.35
    зья
    0.35
     સં
    0.34
    像
    0.34
     semic
    0.34
     حوزه
    0.34
     کھانا
    0.34
    POSITIVE LOGITS
    put
    0.41
     satisfactory
    0.39
    Hammer
    0.39
     reply
    0.38
     replies
    0.38
     คิด
    0.38
    felt
    0.38
     put
    0.37
    𝓰
    0.37
    lag
    0.37
    Activations Density 0.000%

    No Known Activations