© 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. 10-GEMMASCOPE-2-TRANSCODER-262K
    4. 46223
    Prev
    Next
    INDEX
    Explanations

    The neuron activates when a word ending in "ly" is followed by a capitalized word, often a single word. The capitalized word appears to be the start of a specific phrase, title, or important term. The TOP_POSITIVE_LOGITS suggest a wide range of abstract concepts, but the MAX_ACTIVATING_TOKENS and TOKENS_AFTER_MAX_ACTIVATING_TOKEN provide a more concrete pattern.Looking at the pattern:- `MAX_ACTIVATING_TOKENS`: `"`, `"`, `ly`, `"`, `“`, `ly`, `ly`, `ly`, `ly`, `"`- `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: `Lon`, `Stay`, `*`, `Echo`, `Lost`, `***`, `*`, `*.`, `*`, `Razor`Many occurrences of "ly" are followed by capitalized words. The specific examples are:- `ly` -> `Lon` (in `Loneliness:`)- `ly` -> `Stay` (in `Stay down`)- `ly` -> `Echo` (in `Echoes of Elysium`)- `ly` -> `Lost` (in `Lost, little dryling?`)- `ly` -> `Razor` (in `Razor and Blades`)The pattern seems to be: adverb ending in "ly" followed by a capitalized word, which often starts a significant phrase or name. The quotation marks in `MAX_ACTIVATING_TOKENS` also suggest these phrases are often enclosed or emphasized.Let's check the `TOP_POSITIVE_LOGITS`.A word ending in "ly" followed by a capitalized word

    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_10_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
    an
    0.57
    ul
    0.54
    i
    0.54
    ut
    0.52
    י
    0.50
    ان
    0.50
    una
    0.49
    es
    0.47
    um
    0.47
    де
    0.45
    POSITIVE LOGITS
     intermédiaire
    0.48
     răm
    0.46
     získ
    0.46
     argue
    0.45
     intermediary
    0.44
    會
    0.44
     základ
    0.44
     insol
    0.44
     occidentale
    0.44
     intima
    0.44
    Activations Density 0.001%

    No Known Activations