© 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. 4674
    Prev
    Next
    INDEX
    Explanations

    mentions of legal and citizenship-related terms and topics

    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
    27
    +0.11
    0.5%
    646
    +0.10
    0.4%
    314
    +0.09
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    567
    +0.11
    0.05
    646
    +0.10
    0.04
    100
    +0.09
    0.04
    Negative Logits
     Subtle
    -0.77
     scenes
    -0.74
     Atmosp
    -0.74
     louder
    -0.73
     yells
    -0.70
    tone
    -0.68
    lest
    -0.68
    wo
    -0.67
     distract
    -0.67
     Scenes
    -0.67
    POSITIVE LOGITS
     eligibility
    1.58
     exemption
    1.43
     registration
    1.42
     visa
    1.35
     exemptions
    1.34
     eligible
    1.33
     citizenship
    1.31
     visas
    1.31
     paperwork
    1.30
     qualifying
    1.25
    Activations Density 1.152%

    No Known Activations