© 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. 11-GEMMASCOPE-2-TRANSCODER-262K
    4. 91354
    Prev
    Next
    INDEX
    Explanations

    * **ée*** **ée*** **ée*** **Musée*** **ée*** **Musée*** **Musée*** **Musée*** **ée*** **de*** **d*** **Carnaval*** **Matisse*** **d*** **d*** **d*** **du*** **d*** **d*** **inducing*** **v*** **f*** **ı*** **υ*** **ジョ*** **о*** **j*** **const*** **ο*** **Sights:** **Cathédrale Saint-Étienne d'Agde** (a fortress-like cathedral), **Musée de l'Éphèbe** (archaeological museum with a stunning bronze statue discovered in the Mediterranean), ** MUSEUMS & ART (For the Culture Vulture - Allow 2-4 hours each)** * **Musée d'Orsay:** (3-4 hours). Impressionist and Post-Impressionist art in a art galleries, trendy boutiques, and historic Jewish quarter.", "landmarks": ["Place des Vosges", "Musée Carnavalet", "Musée Picasso", "Rue des Rosiers"] }, { "number , and enjoy the sunshine. * **Afternoon:** Choose your art adventure: * **Musée Matisse:** Dedicated to the works of Henri Matisse. * **Musée Marc Chagall:** Focuses on **3. Culture & Arts:** * **Museums:** Beyond the Louvre, Paris boasts incredible museums like the Musée d'Orsay Explore the palace, gardens, and Marie Antoinette's estate. * **Day 4:** Musée d'Orsay (Impressionist art), stroll through the Tuileries Garden, shopping on the Champs-É . It just *feels* special. * **Art & Culture:** World-class museums like the Louvre and Musée d'Orsay, incredible architecture (Notre Dame, Eiffel Tower, Sacré-Cœur), a thriving theater scene **George V:** Yes * **Louvre – Rivoli:** Yes * **Palais Royal – Musée du Louvre:** Yes (some entrances) * **Tuileries:** Yes * **Champs- * the Louvre (though the Mona Lisa is there!). It's a city overflowing with incredible museums. From the Musée d'Orsay (Impressionism!), to the Centre Pompidou (modern art), the Rodin Museum ( **Seine River Cruise:** A great way to see the city from a different perspective. * **Musée d'Orsay:** Art museum housed in a former railway station. * **Disneyland Paris:** Theme parkThe `MAX_ACTIVATING_TOKENS` repeatedly show "ée" and "Mussée".The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` show "de", "d", "du", which are common prepositions following French place nouns like "Musée".The `TOP_ACTIVATING_TEXTS` heavily feature the word "Musée" in contexts referring to specific museums (Musée d'Orsay, Musée Matisse, etc.).The pattern is clearly related to the French word "Musée" and its common grammatical context after the token "ée".Concise explanation needed: 3 to 20 words.It should capture what the neuron detects or predicts.Possible patterns:- Words ending in "ée" followed by prepositions.- French museum names.- French words with accents.Looking at `TOP_POSITIVE_LOGITS`: `inducing`, `v`, `f`, `ı`, `υ`, `ジョ`, `о`, `j`, `const`, `ο`. These are very diverse, including non-Latin characters, making them less helpful for a specific linguistic pattern related to French. They might indicate a general "feature detection" or "token identification" functionality, but not a specific language or concept pattern useful for this explanation.The most concrete and consistent pattern across the relevant lists is the association of "ée" and "Musée" with French museum names.Let's try to form a phrase:- Detecting French museum names. (4 words)- French museum names. (3 words)- Musée followed by prepositions. (4 words)- Words related to French museums. (5 words) - Rule violation: "words related to"Focusing on the core artifact: "Musée" which is preceded by "ée". And the common following prepositions 'de', 'du'.What if the neuron is detecting the characteristic French accent mark 'é'?`MAX_ACTIVATING_TOKENS` shows "ée" multiple times.`TOKENS_AFTER_MAX_ACTIVATING_TOKEN` shows "de", "d", "du".`TOP_ACTIVATING_TEXTS` show "Musée" very frequently.So the pattern is essentially detecting French words ending in 'ée' which are often

    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_11_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
     ن
    0.62
     scan
    0.59
     נ
    0.58
     preposition
    0.58
     n
    0.57
     m
    0.57
     tap
    0.57
     cognac
    0.57
     h
    0.56
     bib
    0.54
    POSITIVE LOGITS
    inducing
    0.63
    v
    0.62
    f
    0.59
    ı
    0.59
    υ
    0.59
    ジョ
    0.58
    о
    0.57
    j
    0.57
    const
    0.56
    ο
    0.55
    Activations Density 0.001%

    No Known Activations