Neuron Explorer

Qwen3-0.6B

comma after 'who' or 'which'

A neuron in layer 8 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

The neuron wakes up on the comma that comes right after words like 'who' or 'which' when they start a side clause, such as 'a role model is someone who, through hard work,'. Turning it up made the model's writing drift into unrelated topics, so the steering results do not show a clear link to this pattern.

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

    the offspring which the LORD will give you by this young woman." (Ruth 4:12).⏎There are certain passages of Scripture which, when you read them, seem out of place with the passages that surround it.

  2. #2 peak

    . Think of a world where you could selectively download your life’s memories- sights, sounds, experiences- onto a Blue-ray to show your grandkids (who, by that time, will have something better than Blue-rays). Imagine being able to

  3. #3 peak

    and/or less comfortable expressing themselves via words. Art Therapy is also known to help people with social skills, specifically, those that are withdrawn or shy, or who, for one reason or another, have a difficult time functioning within social situations.⏎Why

  4. #4 peak

    cards over three rounds, each new round begins after seven cards have been won. This is accomplished by translating cards in both languages. The winner is the player who, in the final round, translates all of the played cards error-free.⏎Novices

  5. #5 peak

    happens when preparation meets opportunity.⏎10. If suitable role models are not nearby, I will seek them out.⏎11. A role model is someone who, through hard work and a positive outlook<|endoftext|>Peace Officers' mourning ribbon⏎The loop

  6. #6 peak

    surcharged with the inspiration of heroism. So the bow was the natural weapon of the simple, perfect physical manhood represented in the idea of Apollo, who, with drawn bow, was the symbol of such manhood displayed in its highest powers

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 Ukrainian president, Viktor Yanukovych, in the black and red flag state of Kyiv, which has a population of 2.4 million, and the population of the Republic of Crimea, which is 4.5 million

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

    the morning of June 28. However, with the atmosphere so dry, the rain associated with the storms evaporated before ever touching the ground, leaving the lightning to ignite the drought-stricken vegetation that hadn’t burned in over 40

  2. #2 peak

    24th October, 1946, when a V-2 rocket lifted the first ever space-bound camera above New Mexico on a short sub-orbital flight. From an altitude of 65 miles, a 35

  3. #3 peak

    tragedy. Excessive heat, prolonged drought, and dry thunderstorms came together in a perfect storm to fuel a raging wildfire that burned 8

  4. #4 peak

    has been checked up to the power 4.6 ⋅<|endoftext|>Picture Yourself in Orbit: Are You Really Weightless?⏎An object in orbit is constantly falling, and falling is what causes “weightlessness.” Gravity acts on you even while

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. L6:2255 weakly excites
  2. L5:1507 weakly suppresses
  3. L2:119 weakly excites
  4. L6:742 weakly suppresses
  5. L6:2575 weakly suppresses
  6. L5:2655 weakly suppresses
  7. L4:2953 weakly excites
  8. L7:3033 weakly excites
  9. L4:2130 weakly excites
  10. L5:1293 weakly excites 'of' in organisation names like 'Ministry of'
  11. L5:640 weakly excites
  12. L2:2431 weakly excites
  13. L0:682 weakly excites
  14. L6:1406 weakly suppresses
  15. L4:397 weakly excites
  16. L4:2986 weakly suppresses
  17. L2:2606 weakly excites
  18. L1:1124 weakly suppresses
  19. L3:2615 weakly suppresses
  20. L5:2058 weakly suppresses
  21. L6:1072 weakly suppresses
  22. L4:708 weakly suppresses
  23. L6:1045 weakly suppresses
  24. L1:2444 weakly suppresses

Feeds later layers

  1. L13:2439 excites
  2. L13:1327 weakly excites
  3. L10:1925 weakly excites
  4. L20:2969 weakly excites
  5. L18:1617 weakly excites
  6. L16:1006 weakly suppresses
  7. L18:244 weakly excites
  8. L14:189 weakly excites 'after' starting a time or cause clause
  9. L15:2502 weakly suppresses the gap in number ranges
  10. L16:2435 weakly excites
  11. L17:3058 weakly excites
  12. L21:278 weakly suppresses
  13. L16:1643 weakly suppresses
  14. L17:2710 weakly excites
  15. L19:2280 weakly suppresses
  16. L9:534 weakly excites
  17. L15:1528 weakly excites
  18. L16:1569 weakly excites
  19. L17:1457 weakly excites
  20. L24:1571 weakly excites
  21. L18:597 weakly suppresses
  22. L25:3028 weakly suppresses
  23. L19:911 weakly excites
  24. L21:2960 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.

3 values between -1.24 and -1.171 values between -1.09 and -1.024 values between -1.02 and -0.942 values between -0.94 and -0.873 values between -0.72 and -0.646 values between -0.64 and -0.579 values between -0.57 and -0.4915 values between -0.49 and -0.4218 values between -0.42 and -0.3443 values between -0.34 and -0.2795 values between -0.27 and -0.19301 values between -0.19 and -0.11623 values between -0.11 and -0.041413 values between -0.04 and 0.04734 values between 0.04 and 0.11350 values between 0.11 and 0.19161 values between 0.19 and 0.2661 values between 0.26 and 0.3429 values between 0.34 and 0.4114 values between 0.41 and 0.493 values between 0.49 and 0.563 values between 0.56 and 0.642 values between 0.64 and 0.712 values between 0.71 and 0.793 values between 0.79 and 0.872 values between 0.87 and 0.941 values between 0.94 and 1.022 values between 1.02 and 1.092 values between 1.09 and 1.171 values between 1.47 and 1.541 values between 1.54 and 1.621 values between 1.69 and 1.77offnegativepositiveoften

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. cut
  2. ?";⏎
  3. ?");⏎
  4. originally
  5. odo
  6. folding
  7. =(⏎
  8. oft
  9. guided
  10. ��

Suppresses

  1. rypton
  2. prox
  3. stellen
  4. reich
  5. داد
  6. phoon
  7. andbox
  8. StateToProps
  9. 考点
  10. Advantage