Neuron Explorer

Qwen3-0.6B

"and" joining clauses after a condition

A neuron in layer 19 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 the word "and" when it joins a second clause that follows an if-clause or a time phrase, such as "If you miss a dose of X and..." or "once the list is shown and...". Turning it down made the model repeat words like "is" and "then", while turning it up gave loose, rambling 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

    5 volumes. If data redundancy on a write-heavy volume is needed, consider mirroring.⏎If the different components in the RAID 5 volume reside on different controllers and the accesses to the volume are primarily large sequential accesses, then setting the interlace

  2. #2 peak

    , therefore, the Bible teaches that polygamy is acceptable.⏎The skeptic seemed to be saying that if a certain behavior is found in the Old Testament narratives, and that behavior is not specifically condemned in those same narratives, then the narratives are teaching

  3. #3 peak

    Let your child know that these cremated remains are placed in a⏎container called an urn and returned to the family. If cremation⏎has already taken place and the container picked up, you may want⏎to show it to the child.

  4. #4 peak

    beautiful flower garden all season long. You can also plant flowers at different stages. Try planting some seeds and some flowers, then by the time your flowers are fading and ready to be<|endoftext|>Protein’s Role in Active & Healthful Aging⏎Earlier

  5. #5 peak

    it also allows for creating a secondary drop-down list that is completely dependent on the first or primary list. In other words, once the first list is displayed, and a parameter chosen, then a subset might exist for any given parameter. By further

  6. #6 peak

    it is growing. This weakens the cell wall and it ruptures resulting in the death of the bacteria.⏎If you miss a dose of Ampicillin and are taking it regularly, take it as soon as possible. If it is almost

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 is is, then 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

    harms the body when damaged cells divide uncontrollably to form lumps or masses of tissue called tumors (except in the case of leukemia where cancer prohibits normal blood function by abnormal cell division in the blood stream). Tumors can grow and interfere

  2. #2 peak

    facets of one and the same underlying structure, in the same manner in which magnetism and the electrostatic force are facets of a more general description of electromagnetism.⏎After Einstein, quite a number of scientists have searched for a unified description of

  3. #3 peak

    monuments that would stand for all time. On a lifeless, airless satellite there would never be any scavengers or souvenir hunters, no wind to bury or wear down the abandoned spacecraft and artifacts, and no air to corrode metal. Even

  4. #4 peak

    should come as little surprise that Olive Ash can be a dead ringer for actual Olive (with the exception of the porous grain structure, which gives its true identity away easily), because both Ash and Olive are placed in the same family: Oleace

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. L17:1048 strongly suppresses
  2. L15:725 strongly suppresses
  3. L16:1174 strongly excites
  4. L16:768 strongly excites
  5. L17:580 strongly suppresses
  6. L16:636 strongly excites
  7. L16:1038 excites
  8. L16:699 excites
  9. L11:237 suppresses
  10. L15:2936 excites
  11. L15:125 excites
  12. L16:1826 excites
  13. L18:1179 suppresses
  14. L15:2505 excites
  15. L14:850 excites
  16. L16:1736 excites
  17. L14:2194 suppresses
  18. L16:669 excites
  19. L7:419 excites
  20. L16:1306 excites
  21. L16:1838 excites
  22. L8:174 excites
  23. L15:806 excites
  24. L14:2932 excites

Feeds later layers

  1. L20:2430 strongly suppresses
  2. L21:881 strongly suppresses comma after 'if' clause in prose, braces in code
  3. L20:2280 strongly suppresses
  4. L24:1443 suppresses
  5. L20:637 suppresses
  6. L20:1799 suppresses
  7. L20:2360 suppresses
  8. L20:1732 suppresses
  9. L20:1312 suppresses
  10. L22:689 suppresses comma closing an intro clause
  11. L20:122 excites
  12. L21:2045 suppresses unclear: weak noun and pronoun activity
  13. L25:1425 suppresses
  14. L24:405 suppresses
  15. L21:1499 suppresses question marks ending a question
  16. L21:809 suppresses
  17. L24:880 suppresses
  18. L20:750 suppresses unclear: chinese academic text, 'when' loops
  19. L27:1964 suppresses
  20. L20:1077 suppresses
  21. L21:1070 weakly suppresses
  22. L21:1805 weakly suppresses
  23. L21:2536 weakly suppresses
  24. L21:930 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 -8.08 and -7.673 values between -6.86 and -6.451 values between -6.04 and -5.634 values between -4.82 and -4.415 values between -4.41 and -4.007 values between -4.00 and -3.5910 values between -3.59 and -3.1921 values between -3.19 and -2.7832 values between -2.78 and -2.3750 values between -2.37 and -1.9665 values between -1.96 and -1.56110 values between -1.56 and -1.15159 values between -1.15 and -0.74364 values between -0.74 and -0.331606 values between -0.33 and 0.07805 values between 0.07 and 0.48271 values between 0.48 and 0.89121 values between 0.89 and 1.3063 values between 1.30 and 1.7144 values between 1.71 and 2.1141 values between 2.11 and 2.5227 values between 2.52 and 2.9321 values between 2.93 and 3.3418 values between 3.34 and 3.7411 values between 3.74 and 4.1514 values between 4.15 and 4.5610 values between 4.56 and 4.979 values between 4.97 and 5.375 values between 5.37 and 5.784 values between 5.78 and 6.191 values between 6.19 and 6.601 values between 6.60 and 7.002 values between 7.00 and 7.411 values between 7.41 and 7.821 values between 7.82 and 8.23offnegativepositiveoften

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. McL
  2. شاء
  3. EDIA
  4. hower
  5. 因为你
  6. caffold
  7. ряд
  8. cred
  9. Murdoch
  10. uido

Suppresses

  1. 则
  2. ,则
  3. 的话
  4. thì
  5. '){⏎
  6. 的話
  7. ')){⏎
  8. ){⏎
  9. )){⏎
  10. 则是