© 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. 3-GEMMASCOPE-2-TRANSCODER-262K
    4. 127046
    Prev
    Next
    INDEX
    Explanations

    "omatic" followed by a diverse set of follow-up tokens, including programming languages and descriptive words like "way" and "expressions," points towards a neuron that recognizes patterns related to words ending in "omatic" and their subsequent context. The "TOP_POSITIVE_LOGITS" list also contains common letters and syllables, reinforcing the idea that this neuron is sensitive to word structures.Given the rule to be specific and avoid general terms, and looking at the combination of 'omatic' and the wide variety of following words (English, Python, way, expressions, punch, Kotlin, Haskell, C), the neuron seems to be picking up on the suffix '-omatic' irrespective of its specific meaning, but rather as a structural element that can appear before many different grammatical or conceptual constructs.However, the instructions state *not* to list all tokens and to find a *pattern*.The MAX_ACTIVATING_TOKENS is *only* 'omatic'.The TOKENS_AFTER_MAX_ACTIVATING_TOKEN are very diverse.The TOP_POSITIVE_LOGITS are individual letters/syllables.Let's re-read the goal: "concise explanation (3 to 20 words) that captures what the neuron detects or predicts by finding patterns in lists.""The explanation could be about tokens following or preceding certain tokens.""The explanation could be about words starting with a sequence.""Avoid simply listing all the tokens. Instead, try to find patterns."The most obvious pattern is the repeated token in MAX_ACTIVATING_TOKENS: `omatic`.The TOKENS_AFTER_MAX_ACTIVATING_TOKEN doesn't show a strong *semantic* pattern, but a *structural* one: a suffix followed by various common word types.Let's look at the TOP_ACTIVATING_TEXTS for clues about the *purpose* or *meaning* of 'omatic' in context.The text mentions:- "most common and idiomatic way"- "Pythonic Code: Using idiomatic Python features"- "idiomatic Kotlin"- "idiomatic way to filter DataFrames"- "idiomatic C++"- "idiomatic punch"- "idiomatic way to do this in JavaScript"- "idiomatic Haskell"Aha!idiomatic

    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_3_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
    л
    0.87
    ﺐ
    0.79
    5
    0.77
    ين
    0.71
    у
    0.71
    ди
    0.67
    ્રી
    0.65
     psicologia
    0.64
    shooting
    0.64
    シンプルな
    0.64
    POSITIVE LOGITS
    i
    1.34
     n
    0.86
     d
    0.78
    ي
    0.78
    י
    0.78
    I
    0.77
     l
    0.75
     mari
    0.75
     t
    0.75
    an
    0.73
    Activations Density 0.000%

    No Known Activations