© 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 EvalsExportsAPI 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. 5859
    Prev
    Next
    INDEX
    Explanations

    mentions or discussions involving authority figures or influential individuals

    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
    60
    +0.11
    0.5%
    235
    +0.10
    0.5%
    761
    +0.09
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    60
    +0.11
    0.04
    761
    +0.10
    0.04
    235
    +0.09
    0.04
    Negative Logits
    press
    -0.66
    Fed
    -0.62
    EMBER
    -0.61
    henko
    -0.58
    pool
    -0.58
    stress
    -0.57
     Supplement
    -0.57
    TG
    -0.56
     advertisement
    -0.56
    Poll
    -0.56
    POSITIVE LOGITS
    stood
    1.65
     whom
    1.28
     disabilities
    1.22
    standing
    1.17
    drawn
    1.14
     regard
    1.14
     regards
    1.09
     impunity
    1.06
     respect
    1.01
    holding
    0.99
    Activations Density 0.104%

    No Known Activations