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

    Neuronpedia

    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 3072
    9. 8-RES_FS3072-JB
    11. 791
    Prev
    Next
    INDEX
    Explanations

    comparative and superlative adverbs such as "more" and "most"

    oai_token-act-pair · gpt-3.5-turboTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 8-res_fs3072-jb
    Configuration
    jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_3072
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    3,072
    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
    102
    +0.18
    0.8%
    55
    +0.13
    0.6%
    497
    +0.12
    0.5%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    102
    +0.18
    0.15
    55
    +0.13
    0.13
    441
    +0.12
    0.09
    Negative Logits
     Flavoring
    -0.94
     Deaths
    -0.85
    rosis
    -0.78
     Answers
    -0.77
    rities
    -0.76
     VIDEOS
    -0.73
    76561
    -0.72
     Landing
    -0.72
    idth
    -0.70
    ilts
    -0.69
    POSITIVE LOGITS
     comfortable
    1.23
     reliant
    1.15
     efficient
    1.14
     uncomfortable
    1.14
     attractive
    1.12
     difficult
    1.09
     susceptible
    1.09
     sensitive
    1.05
     interesting
    1.05
     profitable
    1.05
    Activations Density 1.091%

    No Known Activations