© 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. 16051
    Prev
    Next
    INDEX
    Explanations

    phrases related to legal violations

    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
    139
    +0.11
    0.5%
    243
    +0.10
    0.5%
    705
    +0.09
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    139
    +0.11
    0.04
    608
    +0.10
    0.05
    705
    +0.09
    0.04
    Negative Logits
    izen
    -0.78
     Improvement
    -0.72
     Units
    -0.71
    estones
    -0.67
     improves
    -0.65
     improvement
    -0.65
    cry
    -0.64
     Survivors
    -0.64
     Improvements
    -0.64
     milestones
    -0.63
    POSITIVE LOGITS
     knowingly
    1.17
     careless
    1.14
     improperly
    1.11
     neglect
    1.09
     omission
    1.08
     neglig
    1.06
     willfully
    1.05
     arrog
    1.04
     blindly
    1.04
     negligent
    1.03
    Activations Density 0.366%

    No Known Activations