© 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. 22957
    Prev
    Next
    INDEX
    Explanations

    instances of apologies and expressions of remorse

    oai_token-act-pair · gpt-4o-miniTriggered 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
    274
    +0.12
    0.5%
    288
    +0.12
    0.5%
    759
    +0.10
    0.5%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    274
    +0.12
    0.02
    759
    +0.12
    0.02
    236
    +0.10
    0.02
    Negative Logits
    tein
    -0.75
    weeney
    -0.72
    adj
    -0.70
    wer
    -0.69
    acles
    -0.68
     vantage
    -0.67
    ulhu
    -0.67
    growth
    -0.66
    Downloadha
    -0.66
    apore
    -0.65
    POSITIVE LOGITS
     unres
    1.03
    giving
    0.99
     apology
    0.96
     apologized
    0.90
     apologize
    0.88
     sincerely
    0.86
     apologizing
    0.83
     apologies
    0.82
     acknowled
    0.79
     apologise
    0.79
    Activations Density 0.031%

    No Known Activations