© 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. 2885
    Prev
    Next
    INDEX
    Explanations

    phrases related to politics and controversial figures

    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
    481
    +0.29
    1.4%
    447
    +0.14
    0.7%
    393
    +0.12
    0.6%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    714
    +0.29
    0.04
    595
    +0.14
    0.04
    713
    +0.12
    0.03
    Negative Logits
    Tes
    -0.77
    rb
    -0.73
    hub
    -0.64
    agon
    -0.64
    ibble
    -0.64
    HAM
    -0.63
    Redditor
    -0.63
    FL
    -0.62
     Powered
    -0.61
    bub
    -0.61
    POSITIVE LOGITS
     assorted
    1.23
     finally
    1.10
     possibly
    1.00
     ultimately
    0.97
     downright
    0.96
     optionally
    0.92
     consequently
    0.91
     perhaps
    0.85
     frankly
    0.85
     occasionally
    0.82
    Activations Density 0.132%

    No Known Activations