© 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 12288
    9. 8-RES_FS12288-JB
    11. 9797
    Prev
    Next
    INDEX
    Explanations

    mentions of criminal activities and investigations

    oai_token-act-pair · gpt-3.5-turboTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 8-res_fs12288-jb
    Configuration
    jbloom/GPT2-Small-Feature-Splitting-Experiment-Layer-8/blocks.8.hook_resid_pre_12288
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    12,288
    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
    129
    +0.13
    0.6%
    762
    +0.10
    0.4%
    734
    +0.09
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    762
    +0.13
    0.10
    129
    +0.10
    0.06
    703
    +0.09
    0.05
    Negative Logits
     Canal
    -0.76
     Improvement
    -0.71
     childbirth
    -0.70
     Guinness
    -0.70
     NH
    -0.69
     Quality
    -0.68
    é¾įå
    -0.68
     Presbyterian
    -0.68
    utical
    -0.68
     Shal
    -0.67
    POSITIVE LOGITS
     exploiting
    1.18
     attacking
    1.17
     stalking
    1.13
     infiltr
    1.10
     invading
    1.09
     prow
    1.08
     attack
    1.07
     threatening
    1.07
     hijacked
    1.06
     harassing
    1.06
    Activations Density 1.190%

    No Known Activations