© 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. Joseph Bloom · Feature Splitting for GPT2-Small
    5. GPT2-Small
    7. Residual Feature Splitting 24576
    9. 8-RES_FS24576-JB
    11. 2760
    Prev
    Next
    INDEX
    Explanations

    customer service related text

    oai_token-act-pair · gpt-3.5-turboTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 8-res_fs24576-jb
    Configuration
    jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_24576
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    24,576
    Data Type
    torch.float32
    Hook Point
    blocks.8.hook_resid_pre
    Architecture
    standard
    Context Size
    128
    Dataset
    Skylion007/openwebtext
    Hook Point Layer
    8
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Neuron Alignment
    Index
    Value
    % of L₁
    96
    +0.10
    0.4%
    186
    +0.10
    0.4%
    615
    +0.09
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    96
    +0.10
    0.02
    32
    +0.10
    0.02
    568
    +0.09
    0.01
    Negative Logits
    enberg
    -0.77
    oggles
    -0.75
    ooks
    -0.74
    lahoma
    -0.74
    rams
    -0.74
    orb
    -0.72
    uits
    -0.72
    hift
    -0.68
    oggle
    -0.68
    eton
    -0.67
    POSITIVE LOGITS
     satisfaction
    0.98
     Satisf
    0.97
     service
    0.92
    ele
    0.92
     testim
    0.89
    atical
    0.84
     SERVICE
    0.81
    service
    0.80
     Loyal
    0.76
     feedback
    0.76
    Activations Density 0.027%

    No Known Activations