© 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-3-27B-IT
    3. 16-GEMMASCOPE-2-TRANSCODER-262K
    4. 100094
    Prev
    Next
    INDEX
    Explanations

    the text contains specific terms that appear to be technical or related to specific domains. The presence of "PCs" followed by a comma, "entropy" potentially related to scientific concepts, "Dish" which could be a product or service name, "license" for software or rights, and "base" in a technical context, points towards a specific vocabulary.The `TOP_POSITIVE_LOGITS` also show a mix of languages (German, Polish, Arabic, Slovak) and specific words like "billboard". This suggests the neuron might be sensitive to:* **Technical terms**: "PCs", "entropy", "license", "base"* **Specific entities/products/services**: "Dish", "billboard"* **Multilingual vocabulary**: "erhältlich", "osoby", "قول", "ľudí"Looking at the direct pairings:* `MAX_ACTIVATING_TOKENS`: `PCs` -> `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: `,`* `MAX_ACTIVATING_TOKENS`: `De` -> `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: `composition` (This suggests "decomposition" or similar terms)* `MAX_ACTIVATING_TOKENS`: `entropy` -> `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: `is`* `MAX_ACTIVATING_TOKENS`: `our` -> `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: `and`* `MAX_ACTIVATING_TOKENS`: `Dish` -> `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: `:**` (Implies "Dish:**" or the start of a description)* `MAX_ACTIVATING_TOKENS`: `license` -> `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: `more` (not a strong pattern)* `MAX_ACTIVATING_TOKENS`: `based` -> `TOKENS_AFTER_MAX_ACTIVATING_TOKEN`: `on` (This is a very strong idiom: "based on")The `TOP_ACTIVATING_TEXTS` reinforce these findings:* "powerful PCs" -> followed by `,`* "**Decomposition:** This is..." -> `De` starts decomposition, `is` follows.* "**The Daily Dish:** (Suggests regular content. Availability: Likely taken, but variations possible)" -> `Dish` is followed by a colon and descriptive text.* "purchasing a license more affordable." -> `license` is followed by context about affordability.* "**I. Brain Function & Focus (Crucial for Discipline)** * **Magnesium (Strong):**" -> `our` is not explicitly in this sample, but it's common in lists.* "Based on your health" -> `based` followed by `on`.The `TOP_POSITIVE_LOGITS` are diverse, including foreign words and common words. This suggests the neuron might be sensitive to a broad range of specific terms, possibly including foreign language terms in certain contexts, or even just common words that appear after key terms.Let's consolidate the patterns:1.technical terms and product namesExplanation: technical software and product terms

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    google/gemma-scope-2-27b-it/transcoder_all/layer_16_width_262k_l0_small_affine
    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
     condizioni
    0.45
    cuno
    0.44
    སྐ
    0.43
    ೊಳ
    0.43
    bridges
    0.41
    ार्
    0.41
    基づ
    0.41
    чени
    0.40
    asjon
    0.40
    dys
    0.40
    POSITIVE LOGITS
    1
    0.66
     erhältlich
    0.56
    3
    0.55
     osoby
    0.49
     Zayed
    0.49
    قول
    0.48
    4
    0.46
     billboard
    0.46
     ľudí
    0.45
    躇
    0.45
    Activations Density 0.000%

    No Known Activations