Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    INDEX
    Explanations

    Project Manager, ICA, DFA

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

    rogue

    np_max-act · gemini-2.5-flash-liteTriggered by @ccummings196

    These examples contain various text passages with isolated tokens or short sequences marked as important. The marked elements appear to be random or contextually insignificant tokens (commas, articles, single letters, numbers, words mid-sentence) rather than following any coherent linguistic pattern. The selections do not correspond to grammatically meaningful units, semantic importance, or discourse function. Without a clear pattern across examples, no meaningful linguistic feature or rule can be extracted from these marked selections.

    eleuther_acts_top20 · claude-4-5-haikuTriggered by @ccummings196

    numerical values or quantities embedded within text.

    oai_token-act-pair · claude-4-5-haikuTriggered by @ccummings196
    New Auto-Interp
    Negative Logits
    ill
    0.61
    mj
    0.54
    telephone
    0.54
    adece
    0.53
    illä
    0.53
    iro
    0.52
    pathy
    0.52
    raphe
    0.52
    öp
    0.52
    rimiento
    0.51
    POSITIVE LOGITS
    〕
    0.55
    )。
    0.54
    山
    0.51
    例
    0.51
    ”。
    0.50
     aftermath
    0.49
    。"
    0.49
     transformers
    0.49
     DDR
    0.49
     magistrates
    0.49
    Act Density 0.000%

    No Known Activations