© 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
    3. 31-GEMMASCOPE-2-RES-262K
    4. 197964
    Prev
    Next
    INDEX
    Explanations

    user<MAX_ACTIVATING_TOKENS>1.0whichthisthanorhavetoaboutasand</MAX_ACTIVATING_TOKENS><TOKENS_AFTER_MAX_ACTIVATING_TOKEN>waswithbytheybeenbeenbenewa</TOKENS_AFTER_MAX_ACTIVATING_TOKEN><TOP_POSITIVE_LOGITS>15.991.99X8.9999T5.9910.9919.99D</TOP_POSITIVE_LOGITS><TOP_ACTIVATING_TEXTS>A very good restaurant with excellent food, ambiance and service. The wine list was extensive and not too badly priced. I would definitely recommend this restaurant to anyone who enjoy Italian food. I had the 1.99 lunch special which was tremendous value. I found the food here to be quite good and the service was excellent. This dish was not on the menu but was created specially for me by the chef. It was excellent. Thank you so much. The owner and staff are very friendly and helpful. My wife and I had dinner here on Friday evening. We ordered the 15.99 prix fixe menu which was a great deal and the food was to die for. I had been here before and loved it then and loved it again this time. I have eaten here many times and it is always fantastic. This place is so popular that we couldn't get a table and ended up eating elsewhere. I also recommend the duck and the Tiramisu. My wife loves the Tiramisu. She always gets it. I thought that the food was really good and the service was outstanding.</TOP_ACTIVATING_TEXTS>Explain the neuron above with a word or phrase, not a complete sentence.thoughtThe user wants me to explain the behavior of a neuron with a concise phrase (3-20 words) based on the provided lists. I need to find a pattern.**MAX_ACTIVATING_TOKENS**:'1.0', 'which', 'this', 'than', 'or', 'have', 'to', 'about', 'as', 'and'These are common English words, some are prepositions, conjunctions, or determiners. '1.0' looks like a number.**TOKENS_AFTER_MAX_ACTIVATING_TOKEN**:'was', 'with', 'by', 'they', 'been', 'been', 'be', 'new', 'a'These are also common English words, mostly verbs or articles.Let's try to pair them from the text, if possible.The text mentions:* "prix fixe menu which was a great deal" -> `which` followed by `was`* "I had the 1.99 lunch special which was tremendous value" -> `which` followed by `was`* "My wife and I had dinner here on Friday evening. We ordered the 15.99 prix fixe menu which was a great deal" -> `which` followed by `was`* "My wife loves the Tiramisu. She always gets it. I thought that the food was really good and the service was outstanding." -> `and` followed by `the` (no, `a` is in the list)* "The wine list was extensive and not too badly priced." -> `and` followed by `not` (not in list).* "I have eaten here many times and it is always fantastic." -> `and` followed by `it`.Let's look at the specific pairings given in the lists:* 1.0 -> was (from text: "99.9% pure water was...") -> Doesn't quite fit the provided text. The "1.0" is likely a numerical value.* which -> was (found multiple times in text)* this -> by (from text: "This dish was...") -> doesn't fit. "this" is followed by "dish".* than -> they (from text: "not better than they...") -> doesn't fit.* or -> been (from text: "X or Y have been...") -> doesn't fit.* have -> been (from text: "have been...") -> found.* to -> be (from text: "to be...") -> found.* about -> new (from text: "about new...") -> doesn't fit.* as -> a (from text: "as a...") -> found.* and -> a (from text: "and a...") -> found.**TOP_POSITIVE_LOGITS**:'15.99', '1.99', 'X', '8.99', '99', 'T', '5.99', '10.99', '19.99', 'D'These are mostly prices/numerical values, or single letters that could be product codes or initials. The prices are a strong signal.**TOP_ACTIVATING_TEXTS**:The texts are reviews of a restaurant. prices and descriptions

    np_acts-logits-general · gemini-2.5-flash-lite

    game variations, image hosting

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-27b-pt/resid_post/layer_31_width_262k_l0_medium
    Prompts (Dashboard)
    392,802 prompts, 256 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    drying
    0.51
     cominci
    0.49
     Interessen
    0.47
     Geraldine
    0.45
    entes
    0.43
     adul
    0.42
    การณ์
    0.42
    👒
    0.41
    nocześnie
    0.41
    political
    0.41
    POSITIVE LOGITS
    ユーザ
    0.46
     பயன்படுத்தி
    0.45
    依靠
    0.44
     x
    0.44
     က
    0.44
     निर्धारित
    0.43
     ఉపయోగ
    0.43
     installiert
    0.43
     பயன்படுத்து
    0.43
     takeout
    0.42
    Activations Density 0.001%

    No Known Activations