© 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. Qwen3.5-4B
    3. 15-RES-MATRYOSHKA-65K
    4. 19917
    Prev
    Next
    INDEX
    Explanations

    The common thread across `MAX_ACTIVATING_TOKENS` is the concept of "lists" or collections (`set`, `bank`, `pool`). The `TOKENS_AFTER_MAX_ACTIVATING_TOKEN` often show "of", suggesting items *within* these lists. The `TOP_POSITIVE_LOGITS` are mostly non-English or peculiar tokens, which might relate to specific identifiers or types of items in these lists.Let's try to synthesize:- `list`, `set`, `bank`, `pool` -> collections/groupings.- `list` followed by `of` -> items in a collection.- `classics` in logits -> a category.The neuron seems to be associated with structures that organize items.Proposed explanation: "list of items" or "collection of items".Let's check the rules: "3 to 20 words", "single word, or phrase, or pattern", "no 'list', 'tokens', 'patterns'", "specific"."list of items" is 3 words. It doesn't use forbidden words. It's specific.The `TOP_ACTIVATING_TEXTS` reinforce this:- "list" followed by "of items" (implied by context of "key binding")- "set" (of items)- "bank" (of valves)- "list" (of contacts)- "list" (of trackers)- "list" (of machine bindings)- "pool" (of NTP servers)- "list" (of names/careers)- "List" (of responses)- "list" (of target objects)The pattern is consistently about *members* or *contents* of a defined grouping."list of members""grouping of elements""items in a set""collection contents""list of items" seems to be the most direct and best fit for the input."list of items" is 3 words.It's a phrase.It doesn't use forbidden words like "tokens" or "patterns".It's specific.It doesn't start with disallowed phrases.It doesn't capitalize if not a proper noun.Let's also consider `MAX_ACTIVATING_TOKENS` like `set`, `bank`, `pool`.This suggests it's not *only* about the word "list". It's about any structure that holds multiple items.So, "collection of items" or "members of a group"."set", "bank", "pool", "list" all fit "collection" or "group".The tokens after often imply "of".I will go with "collection of items".Check again:- Concise (3-20 words): "collection of items" is 3 words. (OK)- Single word, or phrase, or pattern: It's a phrase. (OK)- About tokens following or preceding: This is implied by finding the pattern across texts and tokens. (OK)- Words starting with a sequence

    np_acts-logits-general · gemini-2.5-flash-lite
    New Auto-Interp
    Top Features by Cosine Similarity
    Configuration
    decoderesearch/qwen-3.5-saes/qwen-3.5-4b
    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
    onton
    -0.06
    itches
    -0.06
    ä¸ĢæĿ¡æĿ¡
    -0.06
    ró
    -0.06
    äºķçĦ¶
    -0.06
    ä¸Ģæĸ°
    -0.06
    åģ¥åº·çļĦ
    -0.06
    ollen
    -0.05
    ing
    -0.05
    enumerate
    -0.05
    POSITIVE LOGITS
    insk
    0.05
    åĿ
    0.05
    ÙĤÙĩ
    0.05
    å°ij许
    0.05
     اÙĦسÙĬ
    0.05
    註
    0.05
     classics
    0.05
    gi
    0.05
    acha
    0.04
     sill
    0.04
    Activations Density 0.055%

    No Known Activations