© 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
    3. Andy Arditi · Finding Misaligned Persona Features in Open-Weight Models
    5. Qwen2.5-7B-IT
    7. Resid Post - 131k
    9. 15-RESID-POST-AA
    11. 71316
    Prev
    Next
    INDEX
    Explanations

    the neuron detects instruction or requirement phrasing — modal/directive words like "should"/"должен" and nearby specification language demanding what must be included.

    oai_token-act-pair · gpt-5-miniTriggered by @vetterc0
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    andyrdt/saes-qwen2.5-7b-instruct/resid_post_layer_15/trainer_1
    Dataset (Dashboard)
    Various
    Features
    131,072
    Data Type
    float32
    Hook Name
    blocks.15.hook_resid_post
    Architecture
    standard
    Context Size
    1,024
    Dataset
    monology/pile-uncopyrighted
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    🗻
    -0.08
     instr
    -0.07
    .structure
    -0.07
    -----------
    -0.07
     что
    -0.07
     cellar
    -0.07
     Steel
    -0.06
    都是
    -0.06
    才能
    -0.06
    صديق
    -0.06
    POSITIVE LOGITS
    ('<?
    0.07
    _REAL
    0.07
     Remaining
    0.07
     Köln
    0.07
    を得
    0.07
    _ln
    0.07
     xuống
    0.07
    획
    0.07
     пед
    0.06
    比べ
    0.06
    Activations Density 0.200%

    No Known Activations