Neuron Explorer

Qwen3-0.6B

words for friend or colleague

A neuron in layer 20 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the word just before a word for a friend or colleague is expected, such as 'a friend [and] colleague' or 'friends [and] those'. Turned up, the model wrote more about friends, colleagues and relationships, which matches what it is tied to.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    some of his most visionary work in a laboratory built for him in Shoreham exactly a century ago by Stanford White, the Gilded Age architect who was a friend and admirer.⏎And it would be a surprise to almost everyone that the 9

  2. #2 peak

    326).⏎What Individuals Thought of Che Guevara⏎Che Guevara was a controversial figure even within the communist world. People who were his friends and those who were opposed to him held differing opinions on him and had varying interpretations and

  3. #3 peak

    00) was a pioneer researcher who was honored in 1997 by the creation of an annual Negro League Conference named for him. He was also my friend. This is his monumentally important study of how baseball drew the color line

  4. #4 peak

    1864, the new commander of the Division of the Mississippi was William Tecumseh Sherman. He was in the pleasant situation of being a friend and colleague of General Grant, now General in Chief. Grant appears to have modified his

  5. #5 peak

    1570), French Hebraist. Born in Uzès, near Nîmes, Mercier was a pupil of François Vatable, whom he succeeded as professor of Hebrew at the Collège Royal, Paris, in 1

  6. #6 peak

    . It was our 4th year attending the vaunted educational conference and expo, but this year we heard a somewhat distressing refrain from many of the teachers we spoke to who had already adopted 3D printers into their classroom — they weren

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: the first month in a month of a month, I was very very very very very very very very (first month of first month) I was very very very very very very very (first month of first month) ... this is just a

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

    in life, to go to school, and to make a choice in life. I can watch a movie, or I don’t have to. I could hang out with my friends or I could hang out with

  2. #2 peak

    Latin American custom called la cuarentena (“quarantine”). It’s a 40-day postpartum period during which mothers recuperate from labor and bond with their babies. Minutes earlier, her bouncy 7-year-old daughter let

  3. #3 peak

    baby gets adequate tummy time.⏎When the babies start rolling over, tummy time also introduces them to new sleeping positions.⏎This time can also be used to bond with your baby, by introducing flashcards or high contrast cards to infants.⏎You

  4. #4 peak

    to measure—the endpoint turns green to indicate it is glued. Then drag the other end to the other point. The measure tool shows exactly how far apart its end points are.⏎Tip: Because the Measure tool is a shape, it displays in the

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. L15:1649 strongly excites
  2. L17:1162 excites
  3. L19:1161 excites
  4. L15:2656 excites
  5. L18:2661 excites
  6. L17:3040 excites
  7. L11:1259 suppresses coercion, abuse and harassment
  8. L10:1418 suppresses
  9. L18:2471 suppresses
  10. L9:149 suppresses
  11. L15:1108 suppresses
  12. L16:2051 excites
  13. L16:2494 suppresses
  14. L19:3058 excites
  15. L16:2682 suppresses
  16. L18:2997 excites
  17. L17:2451 excites
  18. L14:3057 excites
  19. L11:2938 excites
  20. L17:2925 excites
  21. L17:1031 suppresses
  22. L12:2739 excites
  23. L12:2634 suppresses
  24. L18:2857 excites

Feeds later layers

  1. L21:2758 excites
  2. L21:2879 excites
  3. L22:1296 excites
  4. L21:1503 excites
  5. L22:1822 excites
  6. L26:1728 excites
  7. L22:2742 excites
  8. L24:741 excites
  9. L21:2200 excites
  10. L23:1607 excites
  11. L21:1782 excites
  12. L21:1610 excites
  13. L21:1643 excites
  14. L22:2284 excites
  15. L21:2031 excites
  16. L21:2809 excites
  17. L22:2306 weakly excites obedience and submission
  18. L24:2484 weakly excites
  19. L21:2248 weakly excites
  20. L24:2960 weakly excites
  21. L25:632 weakly excites
  22. L25:1687 weakly excites
  23. L25:2825 weakly excites
  24. L21:2590 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 -8.48 and -8.061 values between -7.64 and -7.221 values between -4.69 and -4.272 values between -4.27 and -3.851 values between -3.01 and -2.593 values between -2.59 and -2.177 values between -2.17 and -1.759 values between -1.75 and -1.3381 values between -1.33 and -0.91555 values between -0.91 and -0.491816 values between -0.49 and -0.071189 values between -0.07 and 0.35117 values between 0.35 and 0.7858 values between 0.78 and 1.2028 values between 1.20 and 1.6212 values between 1.62 and 2.046 values between 2.04 and 2.467 values between 2.46 and 2.883 values between 2.88 and 3.304 values between 3.30 and 3.722 values between 3.72 and 4.141 values between 4.14 and 4.561 values between 4.56 and 4.981 values between 7.08 and 7.501 values between 7.50 and 7.921 values between 7.92 and 8.34offnegativepositiveoften

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. 合作关系
  2. friendship
  3. colleague
  4. →friend
  5. teammate
  6. 朋友
  7. Friendship
  8. neighbour
  9. friendships
  10. friends

Suppresses

  1. alphanumeric
  2. 가지
  3. ocabulary
  4. ريط
  5. oting
  6. culus
  7. 藓
  8. 抖
  9. 蕊
  10. "text