© 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 · Open Source Sparse Autoencoders for all Residual Stream Layers of GPT2-Small
    5. GPT2-Small
    7. Residual Stream
    9. 3-RES-JB
    11. 8760
    Prev
    Next
    INDEX
    Explanations

    words related to negative characteristics or insults, particularly terms like "jerk" and "dick."

    oai_token-act-pair · gpt-3.5-turbo

    pejorative terms and phrases related to insults and offensive language

    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
    emi
    -0.65
     Cavaliers
    -0.64
     Fathers
    -0.63
     Blackhawks
    -0.63
     Hurricanes
    -0.62
     Prophe
    -0.61
    lain
    -0.61
     Conditions
    -0.60
     Clause
    -0.60
    Feature
    -0.59
    POSITIVE LOGITS
     jerk
    1.14
     jer
    1.13
    usalem
    0.99
    ety
    0.89
    weed
    0.85
     twitch
    0.82
    offs
    0.82
    ometer
    0.82
    hattan
    0.81
     reflex
    0.80
    Activations Density 0.014%

    No Known Activations