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

    Neuronpedia

    APIAssistant AxisNEWCircuit TracerNEWSteerSAE EvalsExports Community BlogPrivacy & TermsContact
    1. Home
    2. Andy Arditi · GPT-OSS BatchTopK SAEs
    3. GPT-OSS-20B
    4. Resid Post - 131k
    5. 23-RESID-POST-AA
    6. 6075
    Prev
    Next
    INDEX
    Explanations

    emotions

    np_max-act-logits · gemini-2.0-flash

    the neuron detects system/role/metadata and instruction-like tokens — i.e., blocks of system or assistant prompt text and formatting.

    oai_token-act-pair · gpt-5-miniTriggered by @forsubscription1
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    andyrdt/saes-gpt-oss-20b/resid_post_layer_23/trainer_0
    Dataset (Dashboard)
    Various
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     originally
    -0.12
     Usually
    -0.11
     Using
    -0.11
     Typically
    -0.10
     Actually
    -0.10
     Redwood
    -0.10
     Exactly
    -0.10
     used
    -0.10
     fifteen
    -0.10
     Could
    -0.10
    POSITIVE LOGITS
     emotions
    0.22
     emotional
    0.22
     anxiety
    0.22
     emotion
    0.20
     insecurity
    0.20
     resentment
    0.20
     discomfort
    0.20
     empathy
    0.19
     loneliness
    0.19
     sadness
    0.19
    Activations Density 0.141%

    No Known Activations