© 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
    2. Gemma-3-12B
    3. 24-GEMMASCOPE-2-RES-16K
    4. 7490
    Prev
    Next
    INDEX
    Explanations

    square root undoes the squaring

    np_acts-logits-general · gemini-2.5-flash-lite

    The neuron fires strongly on words ending in “-ed,” i.e. past‐tense or past‐participle forms.

    oai_token-act-pair · o4-miniTriggered by @jyhe0408

    words that complete a phrase expressing contrast or contradiction, particularly in the construction "the thing is that" or similar patterns where an expectation is being set up or challenged.

    oai_token-act-pair · claude-4-5-sonnetTriggered by @jyhe0408

    clauses using the perfect aspect or past-participle constructions, often with auxiliaries (including contractions/negation) and subordinate that-clauses.

    oai_token-act-pair · gpt-5Triggered by @jyhe0408
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-12b-pt/resid_post/layer_24_width_16k_l0_medium
    Prompts (Dashboard)
    392,802 prompts, 256 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     करतो
    0.77
    racting
    0.77
    그래서
    0.74
    られる
    0.73
    出来る
    0.73
    하게
    0.72
    ようになる
    0.71
     येतो
    0.70
    ractic
    0.68
     આવશે
    0.68
    POSITIVE LOGITS
     previously
    1.58
     originally
    1.53
     wcześniej
    1.52
     ранее
    1.48
     давно
    1.45
     already
    1.44
     원래
    1.42
     previamente
    1.41
    previously
    1.40
     Previously
    1.39
    Activations Density 0.754%

    No Known Activations