Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    APICircuit TracerNEWSteerSAE EvalsExportsSlackBlogPrivacy & TermsContact
    © Neuronpedia 2025
    Privacy & TermsBlogGitHubSlackTwitterContact
    1. Home
    2. Andy Arditi · Finding Misaligned Persona Features in Open-Weight Models
    3. Llama3.1-8B-IT (Instruct)
    4. Resid Post - 131k
    5. 11-RESID-POST-AA
    6. 84480
    Prev
    Next
    INDEX
    Explanations

    Casual text

    np_max-act · gemini-2.0-flash

    The neuron spikes on the document’s author-voice or opinion phrases (e.g. “I …,” “it is obvious,” “office,” “working there”), i.e. self-referential statements and subjective commentary.

    oai_token-act-pair · o4-miniTriggered by @xinyanhu8

    office

    np_max-act · gemini-2.5-flashTriggered by @aardvarkkr

    The neuron activates on pronouns, possessive adjectives, and conjunctions when referring to people, often in the context of personal involvement or identity.

    oai_token-act-pair · gemini-2.5-flashTriggered by @aardvarkkr
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    andyrdt/saes-llama-3.1-8b-instruct/resid_post_layer_11/trainer_1
    Dataset (Dashboard)
    Various
    Features
    131,072
    Data Type
    float32
    Hook Name
    blocks.11.hook_resid_post
    Architecture
    standard
    Context Size
    1,024
    Dataset
    monology/pile-uncopyrighted
    Embeds
    IFrame
    Link
    •my list

    No Comments

    Negative Logits
    servers
    -0.07
    FromDate
    -0.06
     -------↵
    -0.06
    _joint
    -0.06
     thêm
    -0.06
    $query
    -0.06
    	pp
    -0.06
    Particles
    -0.06
     chops
    -0.06
    מ
    -0.06
    POSITIVE LOGITS
     Lud
    0.07
     ";"
    0.06
    -dollar
    0.06
     tidak
    0.06
     велик
    0.06
     Gentle
    0.06
     dereg
    0.06
    „P
    0.06
    overe
    0.06
     encuentra
    0.06
    Activations Density 0.103%

    No Known Activations