© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    Natural Language
    Autoencoders
    NEW
    Assistant AxisNEWCircuit TracerUPDATESteerSAE EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Gemma-3-27B-IT
    3. 12-GEMMASCOPE-2-TRANSCODER-262K
    4. 139479
    Prev
    Next
    INDEX
    Explanations

    This neuron seems to be activated by the concept or action of "boiling", or words ending in "-ing" / "-ization" when related to processes or states. The most prominent recurring theme is "boil" and "boiling".The explanation needs to be a phrase of 3 to 20 words.Let's combine the observations:- `boil`, `boiling`, `iling`, `ization` are common in `MAX_ACTIVATING_TOKENS`.- `boiling frog`, `Boil 'Em`, `Boiling:`, `boil the ocean`, `boiling fate`, `boiling`, `boil the ocean`, `boiling`, `boiling` appear in `TOP_ACTIVATING_TEXTS`.- The phrase "boil the ocean" appears multiple times, indicating a nuanced usage of "boil" to mean attempting too much.- "boiling frog" is also present.The core seems to be related to "boiling" as a verb or state, and common phrases using it.Possible explanations:- boiling, boil the ocean- boil and boiling- use of 'boil' and 'boiling'- boiling and related processes- idioms with 'boil'"boil the ocean" is a strong specific pattern that appears multiple times. "boiling" is also central.Let's look for a concise phrase that captures this."boiling or boil the ocean" - 5 words. This describes two distinct patterns seen."boiling, boiling frog, boil the ocean" - 7 words. More specific."boil and boiling phrases" - 4 words.Considering the prompt asks for what the neuron *detects* or *predicts* by finding patterns, and the rules suggest looking for words starting with a sequence, or patterns in tokens following/preceding.The most striking pattern is the word "boiling" and the phrase "boil the ocean".Let's try to generalize. It's about the verb "boil" and its gerund/participle "boiling", and how they are used, especially in idiomatic expressions.Option 1: `boiling, boil the ocean` (captures the literal and idiomatic)Option 2: `the act of boiling` (focuses on the gerund)Option 3: `boiling and related idioms` (broader)Option 4: `boiling, boil the ocean, boiling frog` (includes a specific idiom)Given the strict word count and rule about not listing tokens *too* much, a phrase that captures the *meaning* or *context* of the recurring words is best. "boil the ocean" is a very specific idiom. "boiling" is the core word.Let's analyze again:`MAX_ACTIVATING_TOKENS`: `boil`, `boiling`, `iling` (suffix of boiling)`TOP_ACTIVATING_TEXTS`: `boiling frog`, `Boil 'Em`, `Boiling:`, `boil the ocean`, `boiling fate`, `boiling`.The pattern is clearly around the word "boil" and "boiling".boiling, boil the ocean

    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_12_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
    Appuntamento
    0.51
    ंड
    0.50
    емости
    0.47
     na
    0.47
    চিতে
    0.46
     commissions
    0.46
     arist
    0.46
    อส
    0.46
    pyr
    0.45
    ša
    0.45
    POSITIVE LOGITS
    वाया
    0.51
    ുകൾ
    0.50
    skill
    0.50
    setminus
    0.49
    showthread
    0.49
    checkbox
    0.48
    exercice
    0.46
    ↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵
    0.46
    ↵↵↵↵↵↵↵↵↵↵↵↵↵
    0.45
    وفر
    0.45
    Activations Density 0.002%

    No Known Activations