© 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. 30768
    Prev
    Next
    INDEX
    Explanations

    deliver a joke", "Give the snail a slightly slow, deliberate voice", "I'm going to deliver a joke", "deliver the punchline with timing", "hit notes *just* a little lower", "Give them personalities", "I launch back into", "series of tongue twisters, slightly exaggerated", "deliver the punchline", "you held that high note", "say it dramatically"This neuron seems to activate when describing actions related to performance, delivery, or speaking, often with an emphasis on timing, drama, or specific ways of saying/doing things.Looking at `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` and `TOP_ACTIVATING_TEXTS`:- `deliver` is followed by `a joke` or `the punchline`.- `Say` is followed by `it dramatically`.- `hit` is followed by `notes`.- `launch` is followed by `back into`.- `Give` is followed by `the snail` or `them personalities`.- `held` is followed by `that high note`.The common theme is performative speech or actions, often with a descriptive element of *how* it's done. The phrases often involve verbs like "deliver", "say", "give", "hit", "held", "launch", paired with descriptions of the action or its target.Let's consider the verbs that appear repeatedly or seem prominent: deliver, say, give, hit, held, launch.`delivery`, `joke`, `punchline`, `voice`, `mimic`, `dramatically`, `notes`, `personalities`, `exaggerated`, `held high note`.The phrase "deliver a joke", "deliver the punchline" seems very strong."Give the snail a voice", "Give them personalities"."Say it dramatically"."hit notes"."held that high note"."launch back into".This neuron seems to be about performing or presenting something, often verbally, with specific instructions or descriptions of the performance.A concise phrase could be around 'performing verbally' or 'delivering actions'.Let's look at `TOP_POSITIVE_LOGITS` again. It is quite diverse.delivering performance instructions

    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
    please
    0.40
    chor
    0.38
     conveys
    0.36
    phonic
    0.36
     lively
    0.36
     repent
    0.35
    tune
    0.35
    joy
    0.35
     mene
    0.35
     tragen
    0.34
    POSITIVE LOGITS
     ቀን
    0.41
     сопро
    0.39
     Trần
    0.38
     Акча
    0.38
     Тен
    0.38
     össze
    0.38
    ptidase
    0.38
     ১২শ
    0.37
     oxalate
    0.37
    从事
    0.36
    Activations Density 0.020%

    No Known Activations