© 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. 8-RES-JB
    11. 15002
    Prev
    Next
    INDEX
    Explanations

    phrases related to admissions of guilt or wrongdoing

    oai_token-act-pair · gpt-3.5-turbo

    instances of the word "admitted" followed by what is being confessed

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 8-res-jb
    Configuration
    jbloom/GPT2-Small-SAEs-Reformatted/blocks.8.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.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

    Negative Logits
    oldown
    -0.80
    iculture
    -0.79
    lems
    -0.76
    ï¸ı
    -0.76
    ackle
    -0.76
     Mehran
    -0.76
    tel
    -0.76
    icultural
    -0.76
    lash
    -0.72
    alted
    -0.71
    POSITIVE LOGITS
     wrongdoing
    1.03
     murdering
    0.97
     raping
    0.92
     committing
    0.90
     possessing
    0.90
     conspiring
    0.88
     poisoning
    0.86
     cheating
    0.84
     owning
    0.84
     manipulating
    0.83
    Activations Density 0.086%

    No Known Activations