© 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-3-27B-IT
    3. 15-GEMMASCOPE-2-TRANSCODER-262K
    4. 14844
    Prev
    Next
    INDEX
    Explanations

    The **MAX_ACTIVATING_TOKENS** list is heavily dominated by "backlinks," "links," and "building"/"builds." This strongly suggests the neuron is related to the concept of link building.The **TOP_POSITIVE_LOGITS** list is a bit confusing with non-English characters, but it also contains "Council," "consulting," and "consult." This might suggest a more formal or advice-giving context, but the core signal from MAX_ACTIVATING_TOKENS is about "links" and "building."The **TOP_ACTIVATING_TEXTS** reinforce the "link building" theme with phrases like "Link Building: A Step-by-Step Guide," "acquiring hyperlinks," "link building is the process," "impact search rankings," "backlinks," and "SEO." It also mentions "link schemes" in one context, which is a negative association.The neuron seems to activate when discussing the process or strategies of acquiring links, especially for SEO or website authority. It might also capture negative connotations like "link schemes."Given the options, "backlinks" is a very strong signal. "Link building" itself is also a direct theme. The "TOKENS_AFTER_MAX_ACTIVATING_TOKEN" includes "schemes" which, when paired with links, points towards strategies, good or bad.Let's try to find a concise phrase.- backlinks- link building- building links- seo backlinks- link strategies- link schemes"backlinks" is just one token. "link building" is two words. "building links" is two words. "seo backlinks" is two words. "link strategies" is two words. "link schemes" is two words.Looking at the MAX_ACTIVATING_TOKENS: backlinks, backlinks, backlinks, links, backlinks, link.This is overwhelmingly about backlinks.The TOP_ACTIVATING_TEXTS also heavily feature "backlinks" and "link building."The LOGITS add some "consulting" or "council" flavor, which could imply *how* to do link building or advice about it.If the neuron finds patterns when tokens are *building* or *backlinks*, what follows?- building is- building .- building ,- building )

    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_15_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
     Mois
    0.39
    metik
    0.36
    ெடு
    0.36
    nis
    0.35
     автомобиля
    0.34
     muslin
    0.33
    平
    0.33
    ální
    0.33
     coû
    0.32
    モー
    0.32
    POSITIVE LOGITS
    ڳ
    0.45
     സെക്രട്ടറി
    0.40
    ᠲ
    0.39
     Council
    0.37
     consulting
    0.37
     वगैरह
    0.37
    consult
    0.36
    ษ
    0.36
     इंस्टीट्यूट
    0.36
     COUNCIL
    0.35
    Activations Density 0.021%

    No Known Activations