© 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. 10340
    Prev
    Next
    INDEX
    Explanations

    organizations and proper nouns related to health, safety, and regulation

    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
    373
    +0.14
    0.6%
    234
    +0.10
    0.5%
    487
    +0.10
    0.5%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    752
    +0.14
    0.08
    616
    +0.10
    0.05
    680
    +0.10
    0.05
    Negative Logits
    aturdays
    -0.83
    acca
    -0.83
    reddits
    -0.81
    ihad
    -0.79
     selfies
    -0.76
    letes
    -0.72
     slurs
    -0.70
     beware
    -0.70
     skirts
    -0.70
     haun
    -0.70
    POSITIVE LOGITS
     Institute
    1.17
     Bureau
    1.17
     Registry
    1.17
     Consumer
    1.15
     Association
    1.13
     Economic
    1.11
     Environmental
    1.10
     Commission
    1.08
     Statistical
    1.06
     Legislative
    1.05
    Activations Density 0.237%

    No Known Activations