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

    Neuronpedia

    APICircuit TracerNEWSteerSAE EvalsExportsSlackBlogPrivacy & TermsContact
    © Neuronpedia 2025
    Privacy & TermsBlogGitHubSlackTwitterContact
    1. Home
    2. Apollo Research · Taylor · Identifying Functionally Important Features with End-to-End Sparse Dictionary Learning
    3. GPT2-Small
    4. Residual Similar CE Local
    5. 6-RES_SCL-AJT
    6. 34429
    Prev
    Next
    INDEX
    Explanations

    phrases related to legal issues and emotional outbursts

    oai_token-act-pair · gpt-3.5-turbo

    contractions and possessive 's' indicated by specific typographic characters, notably involving “âĢ”, “Ļ”, and "s"

    oai_token-act-pair · gpt-3.5-turboTriggered by @danbraun
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    neuronpedia/gpt2-small__res_scl-ajt/6-res_scl-ajt
    Prompts (Dashboard)
    12,288 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    46,080
    Data Type
    torch.float32
    Hook Point
    blocks.6.hook_resid_pre
    Architecture
    standard
    Context Size
    128
    Dataset
    apollo-research/Skylion007-openwebtext-tokenizer-gpt2
    Hook Point Layer
    6
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Neuron Alignment
    Index
    Value
    % of L₁
    352
    +0.10
    0.5%
    247
    +0.10
    0.5%
    518
    +0.10
    0.5%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    352
    +0.10
    0.26
    23
    +0.10
    0.25
    154
    +0.10
    0.25
    Negative Logits
    erning
    -0.67
    imen
    -0.60
     Roc
    -0.60
     tremend
    -0.59
    etheless
    -0.58
    Ͻ
    -0.55
     capsule
    -0.54
    issan
    -0.54
    anwhile
    -0.53
     Unic
    -0.53
    POSITIVE LOGITS
    ymes
    0.80
    acca
    0.78
    Minecraft
    0.68
    ¯
    0.65
    kamp
    0.64
    mable
    0.63
     âĢº
    0.62
    ı
    0.59
    cking
    0.59
    abad
    0.59
    Activations Density 5.966%

    No Known Activations