© 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. 1-RES-JB
    11. 3711
    Prev
    Next
    INDEX
    Explanations

    words related to negative or harmful behavior or actions, such as abusive, deceptive, and oppressive

    oai_token-act-pair · gpt-3.5-turbo

    terms associated with abusive or harmful behaviors and practices

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 1-res-jb
    Configuration
    jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.hook_resid_pre
    Architecture
    standard
    Context Size
    128
    Dataset
    Skylion007/openwebtext
    Hook Point Layer
    1
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    ild
    -0.92
    igating
    -0.87
    inen
    -0.86
    igated
    -0.85
    ighed
    -0.85
    igate
    -0.84
    osal
    -0.83
    oleon
    -0.83
    izen
    -0.82
    Downloadha
    -0.80
    POSITIVE LOGITS
     abusive
    1.26
     citiz
    0.88
     behav
    0.83
     behaviour
    0.83
     oppressive
    0.81
     undermin
    0.78
    volent
    0.77
     tendencies
    0.77
     discriminatory
    0.77
     minded
    0.75
    Activations Density 0.021%

    No Known Activations