Neuron Explorer

Qwen3-0.6B

'soon' and 'quickly' before a verb

A neuron in layer 7 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 words like 'soon' or 'quickly' when they come just before the verb in a sentence, as in 'he soon finds' or 'we quickly acknowledged'. Its opposite side fires on pieces of capitalised acronyms such as PROMISE and PROTECT. Turning it up made the text ramble without a clear theme.

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

    find more information by checking your browser's SSL/TLS client results.⏎At Denison, you will initially continue your high school Calculus education, but you will quickly be exposed to some of the more fascinating questions that really excite mathematicians.

  2. #2 peak

    by Mademoisselle Reuter. The lady directress does her best to win over Wiliam by her show of affection and coquetry but he soon finds put that she is engaged to M. Pelet and returns her efforts with

  3. #3 peak

    chairman of the Presidium of the People's Assembly), at first appeared to be carrying on Hoxha's tradition of hard-line policies. But it soon became clear that he was more flexible than

  4. #4 peak

    In the state of nature, you have to keep seeking and finding to survive. A momentary thrill is released when you find what you need, but the feeling soon passes so you go back to seeking. It's dopamine that delivers the pleasure

  5. #5 peak

    located.⏎Young McIntire took to the plains as a duck does to water. He was soon the possessor of a freighting outfit, with four yoke of oxen

  6. #6 peak

    . Hang a long bow and a quiver of arrows conspicuously in your hall or library, and you will soon discover that no exquisite painting or bit of statuary will receive from guests more attention

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 study is a matter of interest. The question is a matter of interest. This is a collection of studies. This is the case in the United States. The question is what? The question is what? The study of the heart.

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

    best to reduce the risk of HIV transmission from infected pregnant women to their babies during pregnancy and breastfeeding while preserving the health of these children and their mothers.⏎The PROMISE (“Promoting Maternal-Infant Survival Everywhere”) study aims to enroll

  2. #2 peak

    Sites in the United States. As part of NIEHS's Superfund Research Program, the Puerto Rico Testsite for Exploring Contamination Threats (PROTECT) is exploring the relationships

  3. #3 peak

    (DASH) Eating Pattern May Reduce Age-Related Cognitive Decline⏎The Dietary Approaches to Stop Hypertension (DASH) diet is often recommended by physicians to people with high blood pressure or pre-h

  4. #4 peak

    University of Arizona faculty members Dan Ferguson, Gregg Garfin, Mónica Ramirez-Andreotta and Connie Woodhouse designed the Transdisciplinary Environmental Science for Society (TESS) Program. The following parts have been prepared:⏎Part 1: Fundamentals

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. L3:1498 suppresses
  2. L5:1745 weakly suppresses
  3. L2:59 weakly excites
  4. L2:796 weakly excites
  5. L6:16 weakly excites
  6. L5:1406 weakly suppresses
  7. L4:876 weakly excites
  8. L0:760 weakly suppresses
  9. L5:110 weakly suppresses
  10. L5:1772 weakly excites
  11. L6:2590 weakly suppresses
  12. L1:1769 weakly excites
  13. L4:56 weakly excites
  14. L5:1670 weakly excites
  15. L1:590 weakly suppresses
  16. L5:46 weakly excites
  17. L3:1577 weakly excites
  18. L4:870 weakly suppresses
  19. L6:7 weakly suppresses
  20. L1:2507 weakly excites
  21. L3:3069 weakly suppresses
  22. L4:1053 weakly suppresses
  23. L5:2936 weakly suppresses
  24. L6:2978 weakly excites

Feeds later layers

  1. L9:450 suppresses
  2. L17:2036 suppresses
  3. L11:1190 suppresses
  4. L8:605 weakly suppresses
  5. L10:2577 weakly suppresses
  6. L18:1307 weakly suppresses
  7. L23:1257 weakly suppresses
  8. L11:859 weakly excites
  9. L21:277 weakly excites
  10. L15:2260 weakly suppresses
  11. L8:2187 weakly suppresses
  12. L8:1532 weakly excites
  13. L9:1079 weakly excites
  14. L12:2991 weakly excites
  15. L12:1063 weakly suppresses
  16. L17:2698 weakly suppresses
  17. L8:384 weakly excites
  18. L12:1853 weakly suppresses
  19. L9:2041 weakly suppresses
  20. L9:2702 weakly excites
  21. L9:1077 weakly suppresses
  22. L14:399 weakly excites
  23. L27:2337 weakly suppresses
  24. L11:2583 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 -3.17 and -3.031 values between -2.47 and -2.331 values between -2.33 and -2.191 values between -2.19 and -2.051 values between -1.91 and -1.771 values between -1.63 and -1.491 values between -1.49 and -1.351 values between -1.35 and -1.213 values between -1.21 and -1.075 values between -1.07 and -0.932 values between -0.93 and -0.796 values between -0.79 and -0.6511 values between -0.65 and -0.5129 values between -0.51 and -0.37153 values between -0.37 and -0.23633 values between -0.23 and -0.091734 values between -0.09 and 0.05975 values between 0.05 and 0.19282 values between 0.19 and 0.3354 values between 0.33 and 0.479 values between 0.47 and 0.612 values between 0.61 and 0.751 values between 0.89 and 1.041 values between 2.30 and 2.44offnegativepositiveoften

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. 星级酒店
  2. Technician
  3. disguise
  4. davran
  5. 也只有
  6. олос
  7. vanilla
  8. zn
  9. 무
  10. .softmax

Suppresses

  1. nodoc
  2. 之心
  3. setError
  4. 片子
  5. itational
  6. 家属
  7. ursion
  8. icina
  9. 爸
  10. 个百分