Neuron Explorer

Qwen3-0.6B

comma closing a list item

A neuron in layer 13 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 comes right after one item in a list, just before the final 'and' or 'or', as in 'depleted resources, natural or man-made disasters'. It is quieter, and sometimes negative, on the word 'and' in 'between X and Y'. When turned up, the model got stuck repeating a sentence, so the steering is not informative.

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

    this product is for A SINGLE USER ONLY. That means that copying or sharing any part of this product with another teacher, grade level, school, department, district,

  2. #2 peak

    the power of your leg and core muscles.⏎There is a connection between tension headaches and stress. Conscious stretching, a healthy diet, regular exercise, and lots of sleep can lessen tension headaches.⏎Similar to how stretching in the morning

  3. #3 peak

    00<|endoftext|>The reasons and urgency that lead a group of people to just up and leave a place are varied: obsolete technology, depleted resources, natural or man-made disasters. But whatever the case, there will always be people

  4. #4 peak

    processes which ultimately result in excess cost, morbidity, readmissions, and mortality.1-4⏎There are two major types of apnea:⏎

  5. #5 peak

    give him/her a chance to correct himself/herself before being sanctioned.⏎Actions that need disciplinary action are absenteeism, poor performance, negligence, insubordination, and attendance-related issues (tardiness). In the hospitality industry, it can

  6. #6 peak

    loss. The ideal candidate for gingivitis treatment with Dr. Boulard is someone with swollen or bleeding gums, separating teeth, pain while chewing or biting, or persistent bad breath.⏎Gum disease can be caused by many things, including

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: I have to be able to access, and use, and then again, again, again, then, then, then, then, then, then, then, then, then, then, then, then, then, then, then

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

    stratovolcano on the east coast of Sicily, close to Messina and Catania. It lies above the convergent plate margin between the African Plate and the Eurasian Plate. It is the tallest active volcano in Europe, currently standing

  2. #2 peak

    the occasional bud from a deciduous tree.⏎White-winged Crossbills are nomadic irruptive migrants, which makes their location and movements difficult to track and predict. When crops of spruce and tamarack cones are low, they

  3. #3 peak

    ocht, the famous and prestigious Eleven Towns Race that takes place in his country. For years Piet has dreamed of proving that he is a skater as brave and strong as Pim Mulier -- but he had never imagined that his test would

  4. #4 peak

    regions use their alarm calls differently to warn of approaching predators, a linguistic analysis by a team of scientists reveals. The study, which appears in the journal Linguistics and Philosophy, reveals that monkey calls have a more sophisticated structure than was commonly thought.⏎

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. L8:131 suppresses
  2. L8:593 suppresses
  3. L12:51 excites
  4. L9:212 excites
  5. L8:70 excites 'and' between stacked superlatives
  6. L12:1235 suppresses commas inside numbers and lists
  7. L10:688 excites
  8. L7:999 suppresses
  9. L11:682 excites gap before a second number
  10. L12:1246 suppresses
  11. L8:185 suppresses
  12. L10:838 excites comma in a list of paired nouns
  13. L11:406 suppresses
  14. L12:1922 suppresses sentence-ending periods in explanations
  15. L9:2010 suppresses
  16. L9:92 suppresses "and" or "or" joining two items
  17. L12:925 suppresses
  18. L12:959 excites
  19. L12:1095 suppresses
  20. L11:1253 excites
  21. L6:427 suppresses
  22. L9:159 suppresses
  23. L12:2071 excites
  24. L11:548 suppresses

Feeds later layers

  1. L17:2179 suppresses
  2. L17:335 excites
  3. L16:242 weakly excites
  4. L14:2610 weakly suppresses
  5. L14:2156 weakly suppresses opening parenthesis after a term
  6. L14:1124 weakly suppresses
  7. L16:2177 weakly suppresses
  8. L18:39 weakly excites
  9. L19:1006 weakly excites
  10. L16:511 weakly suppresses
  11. L19:2194 weakly suppresses first half of hyphenated compound adjectives
  12. L15:64 weakly excites
  13. L17:203 weakly suppresses
  14. L17:1144 weakly suppresses
  15. L14:1857 weakly suppresses
  16. L21:2707 weakly suppresses
  17. L20:560 weakly suppresses
  18. L16:1504 weakly suppresses
  19. L14:1475 weakly suppresses
  20. L20:2 weakly excites
  21. L16:319 weakly excites
  22. L16:2100 weakly suppresses
  23. L14:686 weakly suppresses
  24. L17:2353 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 -3.51 and -3.282 values between -3.06 and -2.831 values between -2.60 and -2.374 values between -2.15 and -1.9210 values between -1.92 and -1.6915 values between -1.69 and -1.4624 values between -1.46 and -1.2451 values between -1.24 and -1.01105 values between -1.01 and -0.78174 values between -0.78 and -0.55370 values between -0.55 and -0.33785 values between -0.33 and -0.101763 values between -0.10 and 0.13322 values between 0.13 and 0.3690 values between 0.36 and 0.5844 values between 0.58 and 0.8131 values between 0.81 and 1.0426 values between 1.04 and 1.2718 values between 1.27 and 1.4913 values between 1.49 and 1.7210 values between 1.72 and 1.959 values between 1.95 and 2.184 values between 2.18 and 2.409 values between 2.40 and 2.635 values between 2.63 and 2.866 values between 2.86 and 3.096 values between 3.31 and 3.543 values between 3.54 and 3.772 values between 4.00 and 4.232 values between 4.23 and 4.452 values between 4.91 and 5.141 values between 5.36 and 5.59offnegativepositiveoften

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. .bin
  2. 对公司
  3. _clr
  4. .ser
  5. ricks
  6. ombat
  7. ertime
  8. : ⏎
  9. ерь
  10. rag

Suppresses

  1. peon
  2. astic
  3. iotic
  4. pliers
  5. 治理体系
  6. pe
  7. px
  8. افي
  9. 迫切
  10. anitize