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

    Neuronpedia

    APIAssistant AxisNEWCircuit TracerNEWSteerSAE EvalsExports Community BlogPrivacy & TermsContact
    © Neuronpedia 2025
    Privacy & TermsBlogGitHubSlackTwitterContact
    1. Home
    2. Llama3.3-70B-IT (Instruct)
    3. 50-RESID-POST-GF
    4. 34358
    Prev
    Next
    INDEX
    Explanations

    mentallyThe user wants a concise explanation (3-20 words) of a neuron's behavior based on provided lists.The explanation should identify a pattern in the lists, specifically focusing on how the neuron might be detecting or predicting something.Let's analyze the provided lists:1. **MAX_ACTIVATING_TOKENS**: `water`, `R`, `email`, `water`, `lobster`, `gold`, `wind`, `C`, `Korean`, `water`, `mobile`, `Islamic` * These seem like specific entities or concepts.2. **TOKENS_AFTER_MAX_ACTIVATING_TOKEN**: `\n`, `for`, `,`, `quality`, `.`, `prices`, `power`, `by`, `cultures`, `,`, `applications`, `tradition` * These are often conjunctions, prepositions, punctuation, or common follow-ups to the activating tokens.3. **TOP_POSITIVE_LOGITS**: `respectively`, `alike`, `etc`, `ê·¸ë¦¬ê³ł`, `serta`, `åĪĨåĪ«`, `/etc`, `combo`, `çŃī`, `igsaw` * These look like various specific strings, some potentially encoded, indicating specific outputs or pairings. They are not immediately helpful in forming a general concept phrase.4. **TOP_ACTIVATING_TEXTS**: * `viruses through air and water\n4. The impact` * `proficiency in using Python and R for data analysis and have` * `sent through an SMS or email, or a code generated` * `such as air and water quality, to help identify` * `shrimp, crab, and lobster. Studies have shown that` * `energy and gold prices? The prices have` * `energy technologies like solar and wind power. These technologies generate` * `help treat hepatitis B and C by reducing the amount of` * `elements from Chinese and Korean cultures, and influences from` * `energy and water, by predicting demand and` * `used to build web and mobile applications, integrate with third` * `a story from Christian and Islamic tradition. It tells of`Let's try to find a common thread or pattern.* **WATER**: Appears multiple times in `MAX_ACTIVATING_TOKENS` and `TOP_ACTIVATING_TEXTS`. What follows? `\n`, `quality`, `,`. So, water related to quality or simply mentioned.* **R**: Followed by `for`. Python and R are programming languages.* **EMAIL**: Followed by `,`. Used for communication.* **LOBSTER**: Preceded by `crab`, `shrimp`. list of entities and concepts

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Comparing With LLAMA3.3-70B-IT @ 50-resid-post-gf
    Configuration
    Goodfire/Llama-3.3-70B-Instruct-SAE-l50/Llama-3.3-70B-Instruct-SAE-l50.pt
    Prompts (Dashboard)
    10,000 prompts, 128 tokens each
    Dataset (Dashboard)
    lmsys/lmsys-chat-1m
    No Configuration Found
    Embeds
    IFrame
    Link
    Not in Any Lists

    No Comments

    Negative Logits
    ÐĴÑĤ
    -0.11
    isay
    -0.10
    'gc
    -0.09
     actionTypes
    -0.09
    _tF
    -0.09
    umer
    -0.08
    tor
    -0.08
    ledo
    -0.08
     addCriterion
    -0.08
    thora
    -0.07
    POSITIVE LOGITS
     respectively
    0.15
     alike
    0.12
     etc
    0.11
     ê·¸ë¦¬ê³ł
    0.10
     serta
    0.10
    åĪĨåĪ«
    0.10
    /etc
    0.10
     combo
    0.09
    çŃī
    0.09
    igsaw
    0.08
    Activations Density 0.225%

    No Known Activations