© 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. 16397
    Prev
    Next
    INDEX
    Explanations

    legal and procedural language related to court appeals and criminal prosecutions

    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
    87
    +0.20
    1.0%
    756
    +0.15
    0.7%
    217
    +0.11
    0.5%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    756
    +0.20
    -0.01
    217
    +0.15
    0.04
    87
    +0.11
    -0.03
    Negative Logits
     Hai
    -0.69
     Beautiful
    -0.69
     Created
    -0.67
     Pik
    -0.63
     Merry
    -0.62
    Beaut
    -0.62
     Mushroom
    -0.61
     Expand
    -0.61
     Dimensions
    -0.61
    Morning
    -0.61
    POSITIVE LOGITS
     revocation
    1.28
     adjud
    1.15
     prosecution
    1.11
     lawfully
    1.07
     recourse
    1.04
     statute
    1.03
     statutory
    1.02
     prosecut
    1.02
     perjury
    1.01
     jurisd
    1.01
    Activations Density 0.753%

    No Known Activations