© 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. 17020
    Prev
    Next
    INDEX
    Explanations

    articles talking about political and economic issues related to global governance and finance

    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
    393
    +0.21
    1.0%
    608
    +0.15
    0.7%
    362
    +0.13
    0.6%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    362
    +0.21
    0.06
    608
    +0.15
    0.06
    393
    +0.13
    0.05
    Negative Logits
     besides
    -0.81
    leeve
    -0.77
    SPONSORED
    -0.73
    recy
    -0.70
    encies
    -0.70
     differs
    -0.69
    rg
    -0.67
     resembles
    -0.67
     signifies
    -0.66
     illustrates
    -0.65
    POSITIVE LOGITS
     aforementioned
    1.36
     ones
    1.04
     infamous
    1.00
     slightest
    0.99
     likes
    0.96
     dreaded
    0.95
     latter
    0.90
     smallest
    0.89
     occasional
    0.86
     entirety
    0.85
    Activations Density 0.235%

    No Known Activations