© 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. 10218
    Prev
    Next
    INDEX
    Explanations

    references to a specific TV show or host named "Hardball"

    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
    496
    +0.14
    0.6%
    46
    +0.11
    0.5%
    730
    +0.09
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    46
    +0.14
    0.03
    730
    +0.11
    0.03
    304
    +0.09
    0.02
    Negative Logits
    uality
    -0.79
    umbn
    -0.79
    uating
    -0.75
    utical
    -0.72
    ript
    -0.69
    uations
    -0.68
    entric
    -0.68
    idated
    -0.68
    uated
    -0.67
    ARDIS
    -0.66
    POSITIVE LOGITS
    ened
    1.14
    ening
    1.02
    coded
    0.99
    cover
    0.97
    ball
    0.96
    dri
    0.93
    disk
    0.93
    wired
    0.92
    core
    0.91
    ness
    0.91
    Activations Density 0.023%

    No Known Activations