© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it
    Neuronpedia
    Jacobian LensNEW
    Natural Language
    Autoencoders
    MAY 26    Assistant AxisJAN 26Circuit TracerAPR 26SteerSAE EvalsExportsinterp-engineNEW Community BlogPrivacy & TermsContact
    1. Home
    3. Under Peer Review · Sparse Autoencoders for Pythia-70M-Deduped
    5. Pythia-70M-Deduped
    7. Residuals
    9. 4-RES-SM
    11. 15293
    Prev
    Next
    INDEX
    Explanations

    instances of the verb "was" in various contexts

    oai_token-act-pair · gpt-4o-mini
    Triggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    ctigges/pythia-70m-deduped__res-sm_processed/4-res-sm
    Prompts (Dashboard)
    32,768 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    Features
    32,768
    Data Type
    torch.float32
    Hook Name
    blocks.4.hook_resid_post
    Hook Layer
    4
    Architecture
    standard
    Context Size
    128
    Dataset
    EleutherAI/the_pile_deduplicated
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Neuron Alignment
    Index
    Value
    % of L₁
    307
    +0.19
    1.1%
    306
    +0.15
    0.8%
    466
    +0.13
    0.7%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    23
    +0.19
    0.10
    289
    +0.15
    0.06
    371
    +0.13
    0.08
    Negative Logits
    Ŀ
    -5.08
    ±
    -4.88
    İ
    -4.68
    Ļª
    -4.65
    ¸
    -4.59
    ħ
    -4.57
    ĻĤ
    -4.55
    °
    -4.50
    į
    -4.50
    Ĥ
    -4.50
    POSITIVE LOGITS
     imperative
    2.64
     time
    2.32
     impossible
    2.32
     easier
    2.30
     possible
    2.30
     difficult
    2.29
     unclear
    2.28
     obvious
    2.18
     understandable
    2.13
     easy
    2.08
    Activations Density 0.588%

    No Known Activations