Neuron Explorer

Qwen3-0.6B

"why" and "reason" before an explanation

A neuron in layer 9 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 lights up on the word "why" (and similar words like "reason" or "for") when the text is about to explain a cause or motive. When it was turned up, the written text drifted into odd, unrelated topics, so the effect on the wording was not clear.

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

    picture?⏎These questions will take their focus to something better. Once they’ve calmed down and appear to be past the issue, ask them what happened and explain why that behavior isn’t helpful or necessary.⏎Tip 2. A simple game of

  2. #2 peak

    Eating Helps You Become Aware of Your Eating Habits and Patterns⏎When you take a minute to think about what you’re eating, how you’re eating, and why you’re eating before you actually do it, your brain becomes

  3. #3 peak

    The disorder is chronic and often lifelong. Although symptoms typically begin in adolescence, they usually do not become problematic until mid- to late adulthood. No one knows exactly why the disorder takes so long to manifest; perhaps as those with hoarding symptoms get

  4. #4 peak

    considerable recognition in the specialist, academic⏎and industrial music communities. For further information, and for more⏎detailed descriptions of the approach taken and the justifications for⏎this approach, the following documents are recommended:⏎D., Lenoir,

  5. #5 peak

    brothers. Younger readers will only understand as much about her situation as they already know. This would be a lovely story for readers who are just starting to question why terrible things happen, but still need some distance from the horrific details of war.⏎

  6. #6 peak

    authors and their mentors are responsible<|endoftext|>Young people do not grow and develop in isolation. They need to know whose they are, from whence they have come, and why they are here. Caring parents, grandparents, and older adults are in the

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: You can use a binary search tree to store a collection of numbers. Is the following statement true for the binary search tree: Let S be the set of all binary search trees. The function f(S) = f(T) where T 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

    touching your mouth, nose, and eyes.⏎The COVID-19 virus spreads mainly through dewdrops of saliva or discharge from the nose when the infected individual sneezes or coughs. Therefore, it is vital for you to practice respiratory manners

  2. #2 peak

    and never heal completely, and weaker scar tissue sometimes replaces muscular tissue in these cases. Obesity, lack of use, heavy lifting, chronic cough, and chronic constipation place additional pressures on the pelvic floor. Genetically weak collagen, and smoking (

  3. #3 peak

    all coronaviruses by:⏎- Cleaning your hands frequently using either soap and warm water or alcohol-based hand sanitizers,⏎- Covering your cough and sneezing into your arm to reduce

  4. #4 peak

    divide this into two doses of 500 mg each. Be sure to also drink six to eight glasses of water throughout the day, to help prevent constipation.⏎Certain groups of people are at a higher risk for calcium deficiency. These groups

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. L4:2818 excites
  2. L7:1241 excites
  3. L8:2299 excites
  4. L7:823 excites
  5. L4:1577 excites
  6. L8:831 suppresses
  7. L3:2434 excites
  8. L7:2535 weakly suppresses
  9. L6:2099 weakly suppresses
  10. L5:2926 weakly excites
  11. L4:2898 weakly suppresses
  12. L6:2004 weakly excites
  13. L2:1689 weakly suppresses
  14. L6:2753 weakly suppresses
  15. L7:2002 weakly suppresses
  16. L2:1184 weakly excites
  17. L8:1863 weakly excites
  18. L7:3049 weakly suppresses
  19. L8:2582 weakly excites
  20. L7:1751 weakly suppresses
  21. L6:832 weakly excites
  22. L7:2543 weakly suppresses
  23. L7:1970 weakly suppresses
  24. L5:1345 weakly suppresses

Feeds later layers

  1. L10:2075 excites
  2. L15:2426 excites
  3. L18:2782 excites
  4. L13:1281 weakly suppresses
  5. L16:2565 weakly suppresses
  6. L17:1676 weakly suppresses
  7. L10:1344 weakly excites
  8. L11:1986 weakly excites
  9. L12:2880 weakly excites words that expect an of, to or for next
  10. L14:1814 weakly excites
  11. L10:2402 weakly suppresses
  12. L16:1676 weakly suppresses
  13. L26:572 weakly excites
  14. L27:2378 weakly excites
  15. L15:3006 weakly suppresses
  16. L10:1870 weakly suppresses
  17. L10:1977 weakly excites
  18. L21:1892 weakly suppresses
  19. L21:2421 weakly suppresses
  20. L16:932 weakly suppresses
  21. L16:2175 weakly excites
  22. L10:1821 weakly suppresses
  23. L21:1711 weakly excites
  24. L21:2275 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 -2.63 and -2.321 values between -2.32 and -2.022 values between -1.71 and -1.414 values between -1.41 and -1.105 values between -1.10 and -0.7927 values between -0.79 and -0.49826 values between -0.49 and -0.182642 values between -0.18 and 0.13321 values between 0.13 and 0.4340 values between 0.43 and 0.749 values between 0.74 and 1.058 values between 1.05 and 1.357 values between 1.35 and 1.663 values between 1.66 and 1.972 values between 1.97 and 2.272 values between 2.27 and 2.581 values between 3.80 and 4.111 values between 4.42 and 4.722 values between 4.72 and 5.031 values between 5.95 and 6.261 values between 7.17 and 7.481 values between 8.09 and 8.401 values between 9.32 and 9.63offnegativepositiveoften

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. ONSE
  2. ONS
  3. rieve
  4. يان
  5. hev
  6. еб
  7. قيد
  8. 逄
  9. 为何
  10. 宸

Suppresses

  1. touched
  2. 复制
  3. 试试
  4. loid
  5. <Character
  6. 拭
  7. 相应的
  8. 尝
  9. Wrestling
  10. bart