© 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. 4455
    Prev
    Next
    INDEX
    Explanations

    political terms and phrases related to government activities and negotiations

    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
    314
    +0.10
    0.5%
    98
    +0.10
    0.5%
    717
    +0.08
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    98
    +0.10
    0.06
    283
    +0.10
    0.02
    516
    +0.08
    0.04
    Negative Logits
     Merchants
    -0.66
     stere
    -0.63
    ggles
    -0.62
    ixt
    -0.61
    edin
    -0.61
     sophistication
    -0.60
    styles
    -0.59
     helper
    -0.58
    Reviewer
    -0.58
     tended
    -0.58
    POSITIVE LOGITS
     2020
    0.94
     2024
    0.90
    Brexit
    0.85
     retribution
    0.83
     2021
    0.82
     Article
    0.81
     impeachment
    0.81
     unilaterally
    0.80
     2019
    0.80
     retaliation
    0.78
    Activations Density 0.609%

    No Known Activations