© 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 EvalsExportsAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Gemma-3-27B-IT
    3. 31-GEMMASCOPE-2-RES-16K
    4. 10798
    Prev
    Next
    INDEX
    Explanations

    thoughtThe user wants a concise explanation (3-20 words) for a neuron's behavior, based on the provided lists.The explanation should capture the essence of what the neuron detects or predicts by finding patterns.I need to avoid specific phrases like "This neuron detects/predicts", "tokens", "patterns".The explanation should be specific, not generic.Capitalization is minimal, except for proper nouns.Let's break down the lists:**MAX_ACTIVATING_TOKENS:**- are- An- 1- represent- to- reached- thought- believe- thought- is**TOKENS_AFTER_MAX_ACTIVATING_TOKEN:**- references- anomaly- ate- apoptosis- the- to- they- to- a**TOP_POSITIVE_LOGITS:**- 可能是 (Chinese: possibly/likely)- hoax- likely- 可能 (Chinese: possibly/likely)- 可能 (Chinese: possibly/likely)- احتمال (Arabic: probability)- 실제 (Korean: real/actual)- prawdopod (Portuguese: probably)- Likely- 実際 (Japanese: actually/in reality)**TOP_ACTIVATING_TEXTS:**- "CPC reference in Entity 1 is likely just the retailer. Therefore, it's highly probable these are references to the *same* 2TB WD Red internal hard drive." (mentions "references", "likely", "probable")- "All-domain Anomaly Resolution Office (AARO), documented a significant number of UAP sightings." (mentions "Anomaly", "UAP")- "Early Assumption radiology report: Mild cardiomegaly. Hypoinflation of the lungs. Right basilar XXXX opacity may represent atelectasis." (mentions "Assumption", "represent", "atelectasis")- "more associated with *necrosis* if it's the primary event. **In summary:** The scenario strongly points to apoptosis, and caspase activation is the key mechanism..." (mentions "apoptosis")- "**He Thought He Was in Asia:** Importantly, Columbus believed he had reached the East Indies (Asia). He never realized he had encountered a new continent." (mentions "Thought", "believed")- "This ice shell is thought to be kilometers thick. * **Subsurface Ocean:** Scientists believe a vast ocean of liquid water exists *beneath" (mentions "thought to be", "believe")- "Scientists believe they undertake a remarkable migration, descending hundreds of meters. They have physiological adaptations..." (mentions "believe", "they")- "highlands, thought to be more anorthositic..." (mentions "thought to be")- "This is *the* classic example." (mentions "classic example")**Identifying Patterns:**1. **Probability/Likelihood:** `TOP_POSITIVE_LOGITS` are filled with words like "likely", "probable", "possibility", "probability", "real", "actual". This strongly suggests the neuron is related to determining likelihood or factual status.2. **Connecting `MAX_ACTIVATING_TOKENS` and `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`:** * `are` -> `references` * `An` -> `omaly` * `represent` -> `apoptosis` * `thought` -> `to` (as in "thought to be") * `believe` -> `to` (as in "believe to be")3. likely to be facts or possibilities

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-27b-it/resid_post/layer_31_width_16k_l0_medium
    Prompts (Dashboard)
    238,145 prompts, 512 tokens each
    Dataset (Dashboard)
    lmsys + oasst1
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
     requieren
    0.47
    Discover
    0.44
     замести
    0.41
     rated
    0.40
     പട്ട
    0.40
    Found
    0.40
     विद
    0.40
    ಪ್ಪು
    0.40
    Requires
    0.40
     разделе
    0.40
    POSITIVE LOGITS
    可能是
    0.61
     hoax
    0.60
     likely
    0.60
    可能
    0.58
     可能
    0.54
     احتمال
    0.54
     실제
    0.53
     prawdopod
    0.53
     Likely
    0.52
    実際
    0.52
    Activations Density 0.296%

    No Known Activations