© 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 · Initial Gemma-2B-IT SAEs
    5. Gemma-2B-IT
    7. Residual Stream
    9. 12-RES-JB
    11. 4331
    Prev
    Next
    INDEX
    Explanations

    technical terms related to a specific software or game

    oai_token-act-pair · gpt-3.5-turboTriggered by @bot

    The neuron detects capitalized named entities/proper nouns (e.g., names of people, places, organizations, and other capitalized terms).

    oai_token-act-pair · gpt-5-miniTriggered by @phamvanluong2003
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    jbloom/Gemma-2b-IT-Residual-Stream-SAEs/gemma_2b_it_blocks.12.hook_resid_post_16384
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    HuggingFaceFW/fineweb
    Features
    16,384
    Data Type
    float32
    Hook Name
    blocks.12.hook_resid_post
    Hook Layer
    12
    Architecture
    standard
    Context Size
    1,024
    Dataset
    Skylion007/openwebtext
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Neuron Alignment
    Index
    Value
    % of L₁
    1343
    +0.25
    0.8%
    906
    +0.15
    0.5%
    1842
    +0.13
    0.4%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    1343
    +0.25
    0.07
    764
    +0.15
    0.06
    981
    +0.13
    0.06
    Negative Logits
     rather
    -0.67
     too
    -0.67
     being
    -0.66
    ,
    -0.65
     over
    -0.65
     that
    -0.65
     ‘
    -0.65
     nearly
    -0.65
     such
    -0.65
     regarding
    -0.65
    POSITIVE LOGITS
     quoique
    1.45
     poichè
    1.41
     Kategor
    1.35
     kosme
    1.33
     Traité
    1.32
     Strukt
    1.29
     Outils
    1.28
     véhic
    1.27
     compréhen
    1.25
     convenable
    1.24
    Activations Density 0.325%

    No Known Activations