© 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. Qwen3.5-4B
    3. 15-RES-MATRYOSHKA-65K
    4. 19564
    Prev
    Next
    INDEX
    Explanations

    **Interpretation of Neuron Behavior****Observation:*** `MAX_ACTIVATING_TOKENS`: rich, perfectly, Bright, simple, warm, simple, wide, intelligent, painted, wide* `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: voice, sculpt, crimson, songs, energy, fellow, assortment, song, faces, exThe `MAX_ACTIVATING_TOKENS` often describe a quality or characteristic, and the `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` often represent something that possesses that quality or is modified by it.* `rich voice`* `perfectly sculpt` (or sculpted)* `Bright crimson`* `simple songs`* `warm energy`* `simple fellow`* `wide assortment`* `intelligent songwriting` (from `TOP_ACTIVATING_TEXTS` context, even though `song` is here)* `painted faces`* `wide expanse` (from `TOP_ACTIVATING_TEXTS` context, `wide` -> `expanse`)The pattern is consistently an adjective followed by a noun it modifies. The adjective signifies a descriptive quality, and the noun is the object or concept being described.**Proposed Explanation:**adjective followed by noun

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    decoderesearch/qwen-3.5-saes/qwen-3.5-4b
    Prompts (Dashboard)
    16,384 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    asca
    -0.06
    å®¶å®¶
    -0.06
    èĬ³åįİ
    -0.06
    ÙĨÙĬع
    -0.06
    eneral
    -0.06
    Tail
    -0.06
    alanya
    -0.06
    prototype
    -0.06
    TAG
    -0.06
    翼翼
    -0.06
    POSITIVE LOGITS
    —at
    0.05
    对人çļĦ
    0.05
    ajan
    0.05
    éĢĤæĹ¶
    0.05
    oder
    0.05
     Locker
    0.05
    rg
    0.05
     locker
    0.05
    辺
    0.04
     hum
    0.04
    Activations Density 0.006%

    No Known Activations