© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    APIAssistant AxisNEWCircuit TracerNEWSteerSAE EvalsExports Community BlogPrivacy & TermsContact
    1. Home
    3. Joseph Bloom · Feature Splitting for GPT2-Small
    5. GPT2-Small
    7. Residual Feature Splitting 49152
    9. 8-RES_FS49152-JB
    11. 37359
    Prev
    Next
    INDEX
    Explanations

    instances of verbal abuse or harassment

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 8-res_fs49152-jb
    Configuration
    jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_49152
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    49,152
    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
    288
    +0.11
    0.5%
    571
    +0.10
    0.4%
    503
    +0.09
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    532
    +0.11
    0.01
    571
    +0.10
    0.01
    503
    +0.09
    0.01
    Negative Logits
    arent
    -0.85
    roxy
    -0.83
    bard
    -0.82
    akeru
    -0.81
    enthal
    -0.81
    alach
    -0.79
    avorite
    -0.79
    xon
    -0.78
    ktop
    -0.75
    uden
    -0.74
    POSITIVE LOGITS
     altercation
    1.04
    ized
    0.98
     verbal
    0.97
    isations
    0.96
    izing
    0.92
    izations
    0.92
    ization
    0.91
     communication
    0.89
     abuse
    0.89
     spar
    0.86
    Activations Density 0.007%

    No Known Activations