© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    Jacobian LensNEW
    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. 9-RES-JB
    11. 20261
    Prev
    Next
    INDEX
    Explanations

    words related to harm or dangerous situations

    oai_token-act-pair · gpt-3.5-turbo

    instances of the word "har" related to harm or hurting, along with specific attention to the word "lar" and other segments that hint at creatures or their descriptors

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 9-res-jb
    Configuration
    jbloom/GPT2-Small-SAEs-Reformatted/blocks.9.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.9.hook_resid_pre
    Architecture
    standard
    Context Size
    128
    Dataset
    Skylion007/openwebtext
    Hook Point Layer
    9
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    CRIP
    -0.74
    urally
    -0.70
    eq
    -0.68
    BOOK
    -0.66
    church
    -0.66
     Dempsey
    -0.64
    UCT
    -0.63
    estern
    -0.61
     Verse
    -0.61
    yk
    -0.60
    POSITIVE LOGITS
    riors
    1.27
    riage
    1.14
    rior
    1.07
    acter
    1.02
    rier
    1.00
    assment
    0.96
    vest
    0.95
    ried
    0.93
    ney
    0.92
    allel
    0.92
    Activations Density 0.102%

    No Known Activations