© Neuronpedia 2026
    Privacy & TermsBlogGitHubSlackTwitterContact
    Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    Jacobian LensNEW
    Natural Language
    Autoencoders
    NEW    Assistant AxisNEWCircuit TracerUPDATESteerSAE EvalsExportsinterp-engineNEWAPI Community BlogPrivacy & TermsContact
    1. Home
    3. Joseph Bloom · Initial Gemma-2B SAEs
    5. Gemma-2B
    7. Residual Stream
    9. 0-RES-JB
    11. 1400
    Prev
    Next
    INDEX
    Explanations

    words related to villains, especially when associated with negative traits or actions

    oai_token-act-pair · gpt-3.5-turboTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    jbloom/Gemma-2b-Residual-Stream-SAEs/gemma_2b_blocks.0.hook_resid_post_16384_anthropic
    Prompts (Dashboard)
    24,576 prompts, 128 tokens each
    Dataset (Dashboard)
    HuggingFaceFW/fineweb
    Features
    16,384
    Data Type
    torch.float32
    Hook Point
    blocks.0.hook_resid_post
    Architecture
    standard
    Context Size
    1,024
    Dataset
    HuggingFaceFW/fineweb
    Hook Point Layer
    0
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Neuron Alignment
    Index
    Value
    % of L₁
    50
    +0.08
    0.3%
    67
    +0.06
    0.2%
    1637
    +0.06
    0.2%
    Correlated Neurons
    Index
    P. Corr.
    Cos Sim.
    253
    +0.08
    0.02
    1520
    +0.06
    0.02
    773
    +0.06
    0.02
    Negative Logits
    <bos>
    -1.51
    
    -0.80
    /***
    -0.79
    /**
    -0.70
    <?
    -0.68
    /*
    -0.67
    /*!
    -0.66
    <?
    -0.62
    public
    -0.60
    protected
    -0.60
    POSITIVE LOGITS
     villain
    1.59
     villains
    1.47
    villain
    1.19
     Villain
    1.14
     affez
    1.05
     sappi
    1.03
     evil
    1.00
     palio
    1.00
     affor
    1.00
     chicco
    0.99
    Activations Density 0.153%

    No Known Activations