© 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. 31-GEMMASCOPE-2-TRANSCODER-262K
    4. 167196
    Prev
    Next
    INDEX
    Explanations

    The neuron is identifying individuals, often described as influential, prominent, or associated with specific professions like writer, engineer, or poet. The context often involves biographical descriptions or introductions to such figures. `A` followed by a professional descriptor or an initial name appears frequently.Looking at:- `MAX_ACTIVATING_TOKENS`: `A`, `influential`, `a`, `British`, `a`, `a`, `a`- `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: `writer`, `most`, `the`, `dem`, `highly`, `el`, `highly`, `researcher`, `prominent`- `TOP_POSITIVE_LOGITS`: `mp`, `remote`, `gl`, `lag`, `гла`, `mat`, `レ`, `map`, `MP`, `glace` (These seem less directly related, perhaps representing foreign names or specific tokens in names/places).This neuron correlates "a" with descriptors of individuals, often those who are influential or have specific professions like writers or researchers.Detailed analysis:The `MAX_ACTIVATING_TOKENS` prominently feature the token "A" and "a" multiple times, along with "influential".The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` include descriptive terms like "writer", "most", "highly", "researcher", and "prominent".The `TOP_ACTIVATING_TEXTS` show examples like:- "**Gustavo Violeta**: Ingeniero y militar." (Engineer and military)- "**A historian, writer, and Greek Catholic bishop.**"- "**one of the most important and influential *cantaores***"- "**Eula Lawrence**, a prominent figure"- "**a prolific and influential demoscener and programmer**"- "**a highly acclaimed Pakistani Urdu poet.**"- "**a British elocution teacher.**"- "**a highly influential French poet**"- "**a researcher who extensively studied and promoted**"- "**a prominent Pakistani lawyer and legal scholar.**"The pattern is very clear: the neuron is activating when it sees the article "a" (or "A") followed by a descriptor for a person, often indicating their profession, status, or influence.Therefore, a concise explanation would be:"a" followed by person descriptor

    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_31_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
    赁
    0.40
     Increasingly
    0.38
    ][:
    0.38
     Fitch
    0.38
    '][:
    0.37
     Amerika
    0.37
    [:
    0.37
     Voltaire
    0.36
    始めた
    0.35
     Leibniz
    0.35
    POSITIVE LOGITS
    mp
    0.43
     remote
    0.40
    gl
    0.39
     lag
    0.38
    гла
    0.38
    mat
    0.37
    レ
    0.37
    map
    0.37
    MP
    0.37
     glace
    0.37
    Activations Density 0.000%

    No Known Activations