Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    APICircuit TracerNEWSteerSAE EvalsExportsSlackBlogPrivacy & TermsContact
    © Neuronpedia 2025
    Privacy & TermsBlogGitHubSlackTwitterContact
    1. Home
    2. Joseph Bloom · Open Source Sparse Autoencoders for all Residual Stream Layers of GPT2-Small
    3. GPT2-Small
    4. Residual Stream
    5. 0-RES-JB
    6. 24400
    Prev
    Next
    INDEX
    Explanations

    words related to deception, confusion, and weakening

    oai_token-act-pair · gpt-3.5-turbo

    concepts related to confusion and deception

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 0-res-jb
    Configuration
    jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.hook_resid_pre
    Architecture
    standard
    Context Size
    128
    Dataset
    Skylion007/openwebtext
    Hook Point Layer
    0
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     stewards
    -0.94
     continuing
    -0.77
     hol
    -0.70
     closing
    -0.68
     ongoing
    -0.67
     scheduled
    -0.67
     raft
    -0.66
     loving
    -0.66
     scaling
    -0.66
     reporting
    -0.65
    POSITIVE LOGITS
     confuse
    2.85
     deceive
    2.38
     intimidate
    1.97
     injure
    1.85
     weaken
    1.79
     accuse
    1.70
     complicate
    1.59
     provoke
    1.57
     assassinate
    1.56
     discredit
    1.55
    Activations Density 0.065%

    No Known Activations