© 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. 17278
    Prev
    Next
    INDEX
    Explanations

    descriptions of online content recommendation and interaction in the context of a social media platform such as Facebook

    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
    87
    +0.24
    1.2%
    459
    +0.10
    0.5%
    1
    +0.09
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    87
    +0.24
    -0.03
    225
    +0.10
    0.05
    1
    +0.09
    0.04
    Negative Logits
     ãĤµãĥ¼ãĥĨãĤ£ãĥ¯ãĥ³
    -0.70
    ajor
    -0.69
     Quart
    -0.66
    anship
    -0.65
     recess
    -0.63
     McDonnell
    -0.62
    ********************************
    -0.62
    constitutional
    -0.62
    emate
    -0.62
     Lieutenant
    -0.62
    POSITIVE LOGITS
     URLs
    1.24
     websites
    1.21
     hasht
    1.12
     clicks
    1.09
     keywords
    1.08
     browsing
    1.08
     Gmail
    1.06
     subreddits
    1.06
     Yelp
    1.06
     spam
    1.05
    Activations Density 0.910%

    No Known Activations