© 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 · Open Source Sparse Autoencoders for all Residual Stream Layers of GPT2-Small
    5. GPT2-Small
    7. Residual Stream
    9. 0-RES-JB
    11. 2150
    Prev
    Next
    INDEX
    Explanations

    All the matches for all the activations of neuron 4 surprisingly contain a mistake as they come with activations set to 0, except for liking which has a 10. The neuron is looking for words related to personal preferences or satisfaction

    oai_token-act-pair · gpt-3.5-turbo

    terms related to preference or approval, specifically "liking" and "dislike."

    oai_token-act-pair · gpt-4o-miniTriggered by @bot
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With GPT2-SMALL @ 0-res-jb
    Configuration
    jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.hook_resid_pre
    Architecture
    standard
    Context Size
    128
    Dataset
    Skylion007/openwebtext
    Hook Point Layer
    0
    Activation Function
    relu
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    Ann
    -0.61
     Termin
    -0.60
     Constable
    -0.59
     represents
    -0.58
    ÂŃ
    -0.58
    Hour
    -0.58
    urs
    -0.57
     Tasman
    -0.57
    Div
    -0.56
     Documentation
    -0.56
    POSITIVE LOGITS
     liking
    4.50
     dislike
    1.80
     hating
    1.36
     craving
    1.29
     tastes
    1.28
     preferring
    1.27
     wanting
    1.21
     liked
    1.20
     Likes
    1.18
     taste
    1.15
    Activations Density 0.014%

    No Known Activations