© 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 · Open Source Sparse Autoencoders for all Residual Stream Layers of GPT2-Small
    5. GPT2-Small
    7. Residual Stream
    9. 3-RES-JB
    11. 6159
    Prev
    Next
    INDEX
    Explanations

    responses indicating very strong negative reactions

    oai_token-act-pair · gpt-3.5-turbo

    emotions related to shock and disapproval, particularly "horrified" and "appalled."

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 3-res-jb
    Configuration
    jbloom/GPT2-Small-SAEs-Reformatted/blocks.3.hook_resid_pre
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    24,576
    Data Type
    torch.float32
    Hook Point
    blocks.3.hook_resid_pre
    Architecture
    standard
    Context Size
    128
    Dataset
    Skylion007/openwebtext
    Hook Point Layer
    3
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    icipated
    -0.76
    sites
    -0.71
    hart
    -0.71
    pmwiki
    -0.71
    plane
    -0.70
     dated
    -0.69
    impro
    -0.66
    improve
    -0.66
    amins
    -0.65
    ages
    -0.65
    POSITIVE LOGITS
     aback
    0.83
     stunned
    0.79
     shocked
    0.78
     onlook
    0.78
     horrified
    0.77
    ĸļ
    0.76
     bystanders
    0.74
     citiz
    0.73
     Parish
    0.67
    DAQ
    0.67
    Activations Density 0.057%

    No Known Activations