© 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. 700
    Prev
    Next
    INDEX
    Explanations

    negation-related words and phrases

    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
    642
    +0.11
    0.5%
    551
    +0.10
    0.5%
    307
    +0.10
    0.5%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    307
    +0.11
    0.09
    551
    +0.10
    0.09
    221
    +0.10
    0.08
    Negative Logits
     Seasons
    -0.67
     Maybe
    -0.64
     lousy
    -0.64
    maybe
    -0.63
    å¥
    -0.62
    lined
    -0.62
    Maybe
    -0.61
     Gone
    -0.61
     snipp
    -0.61
     Stuff
    -0.60
    POSITIVE LOGITS
     necessarily
    1.42
    icably
    1.31
    icable
    1.12
     yet
    1.02
    epad
    0.99
    necess
    0.99
    hin
    0.96
     adequately
    0.93
     hesitate
    0.88
     condone
    0.85
    Activations Density 0.480%

    No Known Activations