Neuron Explorer

Qwen3-0.6B

commas inside lists of nouns

A neuron in layer 16 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 a comma that separates items in a list of nouns, such as 'iOS, Android' or 'politics, social issues'. Turning it up produced text that did not show this pattern, so the steering does not confirm it.

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

    and monitor biomass plant activities and perform maintenance as needed.⏎Sample of reported job titles: Auxiliary Engineer, Auxiliary Operator, Biomass Facilitator, Boiler Operator, Control Room Operator,

  2. #2 peak

    come from their association with other subatomic particles. There is the electron neutrino, muon neutrino and tau neutrino, which are linked to the electron, muon and tau, respectively. The electron is the familiar particle from inside atoms

  3. #3 peak

    implement complex algorithms, and integrate various features seamlessly. By leveraging programming languages such as Java, Swift, or Kotlin, developers can build apps for different platforms like iOS, Android, or cross-platform frameworks. Proper coding practices ensure that apps are efficient,

  4. #4 peak

    like and created a video to demonstrate what it might enable you to do.”⏎Video after the break.⏎Read the rest of this entry »⏎Filed under: Android, General technology, Google, Media, Services, Augmented reality, eyeglasses

  5. #5 peak

    manuals, etc. Topics may include⏎sports, environmental problems, politics, social issues, current events, nature,⏎religion, etc. Discuss what you

  6. #6 peak

    , or engage in fun projects. There are numerous online resources and educational platforms that offer interactive STEM activities suitable for everyone from kindergarten to high school such as Khan Academy, STEMfinity, National Geographic Kids, and more. These activities not only promote critical

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: and then and and then and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and

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

    2010, the study notes. The Census sets the drop off from 7 million jobs in 2006 to 4.5 million in 2009.⏎Part of the decline is that about 27

  2. #2 peak

    1 studies—24 sought to increase cervical cancer screening; six to improve the diagnosis and treatment of premalignant lesions of the cervix; one combined screening, diagnosis and treatment).⏎They found evidence that telephone support with navigation increases the rate of

  3. #3 peak

    They provide examples of how to apply standards relevant instruction across core instruction (Tier 1), secondary intervention (Tier 2), intensive intervention (Tier 3) and for to support students with significant cognitive challenges.⏎Taking early action may be key to

  4. #4 peak

    and is in no way intended to replace human translation. Neither BioOne nor the owners and publishers of the content make, and they explicitly disclaim, any express or implied representations or warranties of any kind, including, without limitation, representations and warranties

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. L15:217 suppresses
  2. L8:131 suppresses
  3. L8:593 suppresses
  4. L15:1459 excites
  5. L12:925 suppresses
  6. L9:2726 excites
  7. L8:70 excites 'and' between stacked superlatives
  8. L9:212 excites
  9. L13:977 suppresses
  10. L14:1335 excites
  11. L6:427 suppresses
  12. L14:1804 suppresses "and" in subject-area lists
  13. L10:838 excites comma in a list of paired nouns
  14. L7:519 suppresses
  15. L6:1084 suppresses
  16. L3:250 excites
  17. L14:2054 suppresses
  18. L9:92 suppresses "and" or "or" joining two items
  19. L14:801 excites 'and' or 'or' joining list items
  20. L7:999 suppresses
  21. L9:630 suppresses
  22. L14:366 excites
  23. L15:1340 excites
  24. L4:57 suppresses

Feeds later layers

  1. L18:1207 suppresses
  2. L18:120 excites
  3. L17:1575 suppresses
  4. L18:1413 suppresses
  5. L20:560 suppresses
  6. L19:659 suppresses
  7. L17:1740 suppresses
  8. L20:1009 suppresses
  9. L17:1259 suppresses
  10. L26:347 suppresses the 're' in self-reliant and similar
  11. L17:1492 suppresses
  12. L17:1546 suppresses
  13. L19:1605 suppresses
  14. L18:957 suppresses
  15. L18:1830 weakly suppresses
  16. L20:852 weakly excites
  17. L19:2552 weakly suppresses
  18. L18:104 weakly suppresses
  19. L19:1394 weakly suppresses
  20. L26:65 weakly excites
  21. L18:2341 weakly suppresses comma before an added clause
  22. L17:181 weakly excites
  23. L25:6 weakly excites
  24. L27:1120 weakly suppresses

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 -6.15 and -5.691 values between -4.79 and -4.341 values between -3.89 and -3.442 values between -3.44 and -2.999 values between -2.99 and -2.5411 values between -2.54 and -2.0932 values between -2.09 and -1.6382 values between -1.63 and -1.18237 values between -1.18 and -0.73776 values between -0.73 and -0.282248 values between -0.28 and 0.17278 values between 0.17 and 0.6265 values between 0.62 and 1.0728 values between 1.07 and 1.5223 values between 1.52 and 1.9820 values between 1.98 and 2.438 values between 2.43 and 2.8818 values between 2.88 and 3.3310 values between 3.33 and 3.782 values between 3.78 and 4.239 values between 4.23 and 4.686 values between 4.68 and 5.135 values between 5.13 and 5.587 values between 5.58 and 6.036 values between 6.03 and 6.493 values between 6.49 and 6.943 values between 6.94 and 7.396 values between 7.39 and 7.841 values between 7.84 and 8.292 values between 8.29 and 8.741 values between 8.74 and 9.191 values between 9.19 and 9.643 values between 9.64 and 10.092 values between 10.55 and 11.001 values between 11.45 and 11.90offnegativepositiveoften

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. pool
  2. rone
  3. net
  4. star
  5. Bloody
  6. phans
  7. ixer
  8. rost
  9. abo
  10. cts

Suppresses

  1. 或其他
  2. или
  3. egral
  4. 和技术
  5. 和社会
  6. 和地区
  7. uptools
  8. 或者其他
  9. /Area
  10. nder