© 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 EvalsExportsinterp-engineNEWAPI Community BlogPrivacy & TermsContact
    1. Home
    2. Olmo-3-1125-32B
    3. 32-RES-BATCHTOPK-131K
    4. 89429
    Prev
    Next
    INDEX
    Explanations

    **TOP_POSITIVE_LOGITS**:* overwhelming* overpower* stronger* overcome* undo* sil (likely part of a word like 'silence', 'silken', etc.* smarter* slick* superior* faster These words all suggest a comparison, a superlative, or a state of being beyond a certain threshold or limit.4. **TOP_ACTIVATING_TEXTS**: * "...proved too much for Senator Brown to overcome." -> `much for` * "...sea floor immediately adjacent to the dock is too great and the sea floor immediately adjacent to the dock is too shallow." -> `great` (implies a limit) * "The realization was too much for James Declan to bear" -> `much for` * "...done too quickly" -> `quickly` * "...India are the clear favourites... and only the familiarity with home conditions could tilt the balance in" -> `are` (implies comparison/status) * "until they become so heavy they fall as rain." -> `heavy` (implies a limit) * "The Ravens are impossible to figure out, but they are also impossible to count out..." -> `impossible` * "...a surge is capable to hit and damage my UPS." -> `capable` * "...these efforts were not enough to stop" -> `enough` * "...exceeded the capacity of the facility" -> `capacity` **Synthesizing the information:** The `MAX_ACTIVATING_TOKENS` show words like "much", "great", "heavy", "impossible", "capable", "enough", "capacity". These often describe a state, a limit, or a degree. The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` often follow with phrases indicating purpose or consequence: "for", "to", "of", "the". The `TOP_POSITIVE_LOGITS` are words like "overwhelming", "stronger", "superior", "faster", suggesting something exceeding a norm or rival. The `TOP_ACTIVATING_TEXTS` confirm these themes: "too much for", "too shallow", "not enough to stop", "too quickly", "impossible", "capable to", "exceeded the capacity". The common thread is about exceeding limits, being more than what is required, or reaching a point of maximum capacity or intensity. Let's try to form a concise phrase. - "Exceeding limits" (2 words) - Captures impossible, enough, capacity. - "More than enough" (3 words) - Captures enough, much, great. - "Beyond a threshold" (3 words) - Captures impossible, enough, capacity, much. - "Surpassing a limit" (3 words) - Similar to above. - "Maximum state reached" (3 words) - Captures capacity, heavy. - "Degree or limit reached" (4 words) - A bit too descriptive. - "Intensity or capability" (3 words) - Captures strong, capable, much. The `TOP_POSITIVE_LOGITS` like "overwhelming", "stronger", "superior", "faster" suggest a comparative aspect, being *more* than something else, or reaching an extreme

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    bcywinski/Olmo-3-32B-Base-SAE/saes_allenai_Olmo-3-1125-32B_batch_top_k/resid_post_layer_32
    Prompts (Dashboard)
    16,384 prompts, 128 tokens each
    Dataset (Dashboard)
    monology/pile-uncopyrighted
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    hos
    -0.13
    ysz
    -0.12
     brun
    -0.12
    lena
    -0.12
    ()){
    -0.11
    Saint
    -0.11
    viso
    -0.11
     dared
    -0.11
    isme
    -0.11
     saint
    -0.11
    POSITIVE LOGITS
     overwhelming
    0.10
     overpower
    0.10
     stronger
    0.10
     overcome
    0.10
     undo
    0.09
     sil
    0.09
     smarter
    0.08
     slick
    0.08
     superior
    0.08
     faster
    0.08
    Activations Density 0.035%

    No Known Activations