© 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. OpenAI · Sparse Autoencoder for GPT2-Small - v5 2024
    5. GPT2-Small
    7. Residual Post - 32k
    9. 8-RES_POST_32K-OAI
    11. 4759
    Prev
    Next
    INDEX
    Explanations

    negative forms of the verb "to do" in various contexts

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 8-res_post_32k-oai
    Configuration
    jbloom/GPT2-Small-OAI-v5-32k-resid-post-SAEs/v5_32k_layer_8.pt
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    Skylion007/openwebtext
    Features
    32,768
    Data Type
    torch.float32
    Hook Name
    blocks.8.hook_resid_post
    Hook Layer
    8
    Architecture
    standard
    Context Size
    64
    Dataset
    Skylion007/openwebtext
    Activation Function
    topk
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Neuron Alignment
    Index
    Value
    % of L₁
    373
    +0.12
    0.6%
    281
    +0.10
    0.5%
    317
    +0.09
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    551
    +0.12
    0.06
    49
    +0.10
    0.06
    566
    +0.09
    0.06
    Negative Logits
     refuses
    -0.64
     ancest
    -0.63
    osponsors
    -0.63
     integrity
    -0.60
     constit
    -0.59
     resists
    -0.59
     fundament
    -0.59
     legality
    -0.58
     stances
    -0.58
     opposes
    -0.58
    POSITIVE LOGITS
    croft
    0.78
    IRD
    0.74
    onde
    0.73
    ird
    0.67
    atch
    0.66
    avin
    0.66
    uld
    0.64
    attering
    0.63
    arde
    0.60
    arte
    0.60
    Activations Density 0.284%

    No Known Activations