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

    Neuronpedia

    APICircuit TracerNEWSteerSAE EvalsBlogSlackPrivacy & 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. 8-RES-JB
    6. 56
    Prev
    Next
    INDEX
    Explanations

    the word "published" and phrases related to information being provided.

    oai_token-act-pair · gpt-4oTriggered by @johnny

    words related to the publication or provision of content, especially terms like "published," "provided," and "reproduced."

    oai_token-act-pair · gpt-4.1-mini-2025-04-14Triggered by @johnny

    instances where something is published or provided, typically in the context of distribution of information or materials.

    oai_token-act-pair · gpt-4.1-2025-04-14Triggered by @johnny

    The neuron flags phrases that indicate content is being formally released or made available (e.g. publication or provision statements).

    oai_token-act-pair · o4-miniTriggered by @johnny

    It fires for passive‑voice past participle verbs that describe distributing, releasing, or otherwise making content available.

    oai_token-act-pair · o3Triggered by @johnny

    The pattern involves important words that are past participles or adjectives derived from verbs, often indicating an action related to publishing, providing, reproducing, making, supporting, or translating.

    eleuther_acts_top20 · gpt-4.1-nanoTriggered by @johnny

    mentions of publication sources or identifiers.

    oai_token-act-pair · gpt-4.1-nanoTriggered by @johnny

    verbs related to publishing and providing information

    oai_token-act-pair · gpt-3.5-turbo

    words related to publication and dissemination of content

    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
    Shift
    -0.68
     unlaw
    -0.64
    Operation
    -0.61
     Blacks
    -0.60
    alon
    -0.59
    dor
    -0.58
    tra
    -0.58
    knife
    -0.58
    ua
    -0.57
    Storm
    -0.57
    POSITIVE LOGITS
     by
    0.82
    ĸļ
    0.80
    inently
    0.79
    ependent
    0.77
     chiefly
    0.77
     geographically
    0.76
    ivating
    0.75
    entious
    0.75
     principally
    0.73
     aback
    0.73
    Activations Density 0.301%

    No Known Activations