© 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
    3. Andy Arditi · Finding Misaligned Persona Features in Open-Weight Models
    5. Llama3.1-8B-IT
    7. Resid Post - 131k
    9. 11-RESID-POST-AA
    11. 24942
    Prev
    Next
    INDEX
    Explanations

    motive

    np_max-act · gemini-2.0-flash

    instances of determining or speculating about the motive or cause behind a crime or action.

    oai_token-act-pair · gpt-4oTriggered by @tcai

    Same activations, but with all zeros filtered out: <start> robbery 0.3276 : 0.2369 determine 0.2305 a 0.7637 motive 2.1387 for 1.2744 the 0.7173 shooting 0.3594 , 0.4402 . 0.2903 This 0.2578 happened 0.4341 MB 0.6084 this 0.2445 was 0.9663 an 0.8892 isolated 0.8862 incident 0.7539 and 0.4282 haven’t 0.4282 found 0.9287 a 0.7637 motive 2.1152 for 0.9297 the 0.3113 shooting 0.3594 . 0.2323 Arundel 0.2871 responsible 0.2175 for 0.2109 the 0.3066 attack 0.5010 Arundel 0.2646 investigating 0.3206 a 0.7637 motive 2.0977 for 1.2744 the 0.7173 shooting 0.4343 <end> Explanation of neuron 4 behavior: the main thing this neuron does is find crime-reporting or investigative language cues (mentions of crime acts, incidents, motives and investigation terms).

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

    events related to violent crimes involving multiple victims.

    oai_token-act-pair · gpt-4o-miniTriggered by @xinyanhu8
    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
    Not in Any Lists

    No Comments

    Negative Logits
     VIC
    -0.07
     갈
    -0.07
     aquarium
    -0.07
     Greek
    -0.06
    Greek
    -0.06
    華
    -0.06
    SERVICE
    -0.06
    urations
    -0.06
     ماد
    -0.06
     bedtime
    -0.06
    POSITIVE LOGITS
    \Catalog
    0.07
    /power
    0.06
    .Client
    0.06
    uyệt
    0.06
    ,''
    0.06
     식
    0.06
    订单
    0.06
    ................................
    0.06
    dataType
    0.06
    postId
    0.06
    Activations Density 0.027%

    No Known Activations