© 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. 37-GEMMASCOPE-2-TRANSCODER-262K
    4. 134491
    Prev
    Next
    INDEX
    Explanations

    The neuron likely activates when referring to events or concepts associated with the **current year**.**Reasoning:**1. **TOP_POSITIVE_LOGITS:** This list is overwhelmingly composed of phrases meaning "this year" in various East Asian languages (Chinese, Japanese, Korean) and Indian languages (Marathi). This is the strongest signal for what the neuron is "about".2. **MAX_ACTIVATING_TOKENS:** Tokens like "states", "predict", "Developing", "find", "Extract" often appear in sentences describing activities, plans, or findings.3. **TOKENS_AFTER_MAX_ACTIVATING_TOKEN:** The prevalence of "the" and "a" suggests these MAX_ACTIVATING_TOKENS often precede nouns or noun phrases.4. **TOP_ACTIVATING_TEXTS:** While the texts are diverse, they often discuss plans, strategies, goals, announcements, or updates ("Committee's Purpose", "Initiative Announcement", "main focus for the year", "lead member", "developing a new marketing strategy", "strategic planning", "update your knowledge", "latest news headlines", "Q3 submissions"). The concept of "this year" ties many of these together as the timeframe for these activities.The neuron seems to be finding patterns related to descriptions of actions, plans, or information pertinent to the current year. "This year" is the most direct and specific concept derived from the positive logits

    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_37_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
     tanh
    0.36
     වෙත
    0.34
    딕
    0.33
     Multi
    0.32
     Alternatively
    0.31
    람
    0.31
    ன்கள்
    0.30
    氈
    0.30
     चम्मच
    0.30
    を使用した
    0.30
    POSITIVE LOGITS
    今年的
    1.60
    今年
    1.56
    今年の
    1.48
     এবারের
    1.45
     올해
    1.44
     यंदा
    1.43
     이번
    1.38
    今年は
    1.34
     今年
    1.34
    이번
    1.28
    Activations Density 0.047%

    No Known Activations