Neuron Explorer

Qwen3-0.6B

opening parenthesis before a gloss

A neuron in layer 26 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on an opening bracket that starts an explanation or translation right after a term, like 'virus (or Trojan)' or 'crust (skin)'. Turning it up made the output break into scrambled, mixed-language text.

Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).

Fires most on

The text windows where this neuron's activation was highest. Amber marks positive activation; the outlined token is the peak. Hover a token for its strength.

  1. #1 peak

    ism rather has a different emphasis, shifting the focus of interest toward communities and societies and away from the individual. The question of priority (individual or community) often has the largest impact in the most pressing ethical questions:

  2. #2 peak

    State’s Institute for Behavioral Medicine Research. “They appear to be strongly motivated to try to engage in male sexual behavior with other females.”⏎The researchers found that estrogen (which plays a major role in development of masculine traits in rats) activates mast cells

  3. #3 peak

    extinctions in the past -- the Ordovician, the Devonian, the Permian, the Triassic and the Cretaceous -- the current extinction event (the Anthropocene<|endoftext|>Adolescent Sexual Health:

  4. #4 peak

    and accept aging and the most apparent effect arrives in the form of visible wrinkles. A fine line or crease is observed on the surface of the epidermis (skin). Even though some age more gracefully than the others, the problem continues in

  5. #5 peak

    can withstand water submersion. Although “waterproof” does not have a technical definition, most companies who use the term have evaluated the product’s ingress protection (IP) rating in a laboratory (“ingress” means “entrance” and

  6. #6 peak

    art, perhaps, from hoaxes). However, if you use an HTML-based e-mail client, it can no longer be said that you cannot catch a virus (or Trojan) from an e-mail.<|endoftext|>The coronavirus pandemic has tested our healthcare

Steering

Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.

turned down a lot
Prompt

Here is something I have been thinking about lately: is it really a fact about the is in the is in the is is is is is is is is is is is is is is is is is is is is is is is is is is is is is is is is is is is

Most negative on

Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.

  1. #1 peak

    , most recognizably in the Hebrew name of The Jewish Agency – הַסּוֹכְנוּת הַיְּהוּדִית, or simply

  2. #2 peak

    Continuous⏎помогите пожаааалуйста, сам не справляюсь(⏎Y. Раскройте скобки, употребляя глаголы в Present, Past, Future Indefinite,

  3. #3 peak

    Я РАБОТА ПО АНГЛИЙСКОМУ ЯЗЫКУ⏎КОНТРОЛЬНАЯ РАБОТА ПО АНГЛИЙСКОМУ ЯЗЫ

  4. #4 peak

    ментарии при наличии в них нецензурной брани, оскорблений или спама<|endoftext|>Conflict and Solidarity in Mudejar Society

Connections

Show on map

Which neurons in earlier layers feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.

Fed by earlier layers

  1. L11:223 strongly excites
  2. L18:178 strongly excites
  3. L13:44 strongly suppresses inside web addresses and citations
  4. L15:184 strongly excites
  5. L19:68 excites full stops closing a claim
  6. L10:155 excites
  7. L7:92 excites
  8. L23:359 excites
  9. L15:59 suppresses
  10. L15:390 excites
  11. L18:66 suppresses
  12. L25:30 suppresses
  13. L21:583 excites
  14. L19:1240 suppresses
  15. L22:14 excites
  16. L16:899 excites
  17. L12:158 suppresses
  18. L20:59 suppresses
  19. L19:2829 excites
  20. L17:473 suppresses
  21. L10:669 suppresses
  22. L21:846 excites
  23. L22:349 excites
  24. L20:296 suppresses

Feeds later layers

  1. L27:6 suppresses
  2. L27:19 excites
  3. L27:258 excites
  4. L27:20 excites
  5. L27:59 excites full stops in running text
  6. L27:665 excites
  7. L27:43 excites
  8. L27:94 excites
  9. L27:643 excites
  10. L27:341 excites
  11. L27:481 excites
  12. L27:338 suppresses
  13. L27:265 suppresses
  14. L27:23 suppresses
  15. L27:936 excites
  16. L27:1637 excites
  17. L27:193 suppresses
  18. L27:1490 excites
  19. L27:269 weakly excites
  20. L27:68 weakly excites
  21. L27:21 weakly excites
  22. L27:52 weakly excites
  23. L27:344 weakly excites
  24. L27:17 weakly excites

Activation histogram

How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.

1 values between -11.54 and -10.931 values between -6.09 and -5.481 values between -5.48 and -4.881 values between -4.88 and -4.271 values between -3.66 and -3.06174 values between -3.06 and -2.451099 values between -2.45 and -1.851432 values between -1.85 and -1.24544 values between -1.24 and -0.63145 values between -0.63 and -0.03110 values between -0.03 and 0.5890 values between 0.58 and 1.1853 values between 1.18 and 1.7949 values between 1.79 and 2.4047 values between 2.40 and 3.0033 values between 3.00 and 3.6131 values between 3.61 and 4.2123 values between 4.21 and 4.8220 values between 4.82 and 5.4314 values between 5.43 and 6.0314 values between 6.03 and 6.644 values between 6.64 and 7.247 values between 7.24 and 7.853 values between 7.85 and 8.453 values between 8.45 and 9.061 values between 9.06 and 9.673 values between 9.67 and 10.272 values between 10.27 and 10.881 values between 11.48 and 12.091 values between 12.09 and 12.70offnegativepositiveoften

Logit lens

The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layers instead.

Promotes

  1. 以及其他
  2. than
  3. 那些
  4. 都不能
  5. 的地
  6. supply
  7. 管理和
  8. =post
  9. 但是如果
  10. 企业和

Suppresses

  1. is
  2. over
  3. freely
  4. before
  5. seeks
  6. in
  7. goes
  8. leads
  9. gets
  10. with