Neuron Explorer

Qwen3-0.6B

"of" after a noun introducing a category

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

Fires on the word "of" when it links a noun to what it belongs to, like "the religion of" or "the state tree of". Turning it up made the model write in a loose, wandering style; turning it down made it drift into maths word problems.

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

    cond 'snapshots' reveal a dramatic bond tightening in photo-excited gold complexes⏎More information: "From Binary<|endoftext|>Christianity is, officially, the religion of one third of the people in the world. The religion has spread across the world

  2. #2 peak

    ot observed more than a century ago. S. I. Hayakawa, the semanticist and, later, senator from California, called slang “the poetry of everyday life.” On the other hand, Ambrose Bierce defined slang as “

  3. #3 peak

    00 lifetime risk of contracting the disease, with the average age of detection being between 60 and 70 years.⏎“Cancer is a disease of the aging,” said Dr. Deborah Axelrod, associate professor at New York University

  4. #4 peak

    , if caught early it is easily treatable. Unfortunately, other types of prostate cancer are aggressive and can spread quickly.⏎Prostate cancer is a cancer of the prostate gland. The prostate gland is a walnut-sized gland present only in men

  5. #5 peak

    and reference site. Goethe Verlag has 100 free lessons for Hebrew language learners.⏎Is Hebrew a dead language?⏎Modern Hebrew is the official language of the State of Israel, while premodern Hebrew is used for prayer or study in

  6. #6 peak

    and attractive foliage.⏎Other People Are Reading⏎Eastern redbud's (Cercis canadensis) adaptability has earned it recognition as the state tree of Oklahoma. The quickly growing tree tolerates well-drained clay,

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 area of a triangle is 10 cm. The area of a rectangle is 10 cm? No. So this. So this. So this. So this. So this. So this. So this. So this.

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

    Bottom line is that you can’t go wrong with making blueberries a staple in your grocery cart.⏎Since blueberries are a fruit that are often sprayed quite generously with pesticides,

  2. #2 peak

    watering:⏎The Bucket Method⏎The bucket your tree came in holds about 10 gallons of water. You can use it to water your tree. Line it with a trash bag

  3. #3 peak

    . Troy L. Best is a professor of biological sciences at Auburn University.<|endoftext|>Snare drums are a type of drum that makes a distinct cracking sound when hit with a drumstick. They are also the type of drum on which the drum roll

  4. #4 peak

    the University of Bonn in Germany and colleagues. The team created a three-dimensional interference pattern by shining a laser into a ‘holographic crystal’, and used the pattern as a stencil

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. L5:2937 weakly suppresses
  2. L4:2544 weakly excites
  3. L4:2732 weakly suppresses
  4. L5:117 weakly excites
  5. L2:3024 weakly excites
  6. L5:2873 weakly excites
  7. L5:2497 weakly excites
  8. L5:1158 weakly suppresses
  9. L3:870 weakly suppresses
  10. L2:1266 weakly suppresses
  11. L3:2060 weakly excites
  12. L3:952 weakly suppresses
  13. L1:2844 weakly suppresses
  14. L3:1055 weakly excites
  15. L3:2097 weakly suppresses
  16. L2:2117 weakly suppresses
  17. L0:2896 weakly excites
  18. L5:2582 weakly excites
  19. L5:760 weakly suppresses
  20. L6:1141 weakly excites
  21. L1:543 weakly excites
  22. L2:2576 weakly excites
  23. L0:2122 weakly excites
  24. L0:1975 weakly excites

Feeds later layers

  1. L12:2849 weakly suppresses '-based' after a place name
  2. L12:1906 weakly suppresses
  3. L11:437 weakly excites
  4. L8:2206 weakly excites
  5. L21:1886 weakly excites
  6. L9:955 weakly excites
  7. L8:1039 weakly excites
  8. L24:862 weakly excites
  9. L20:2591 weakly excites
  10. L17:1484 weakly suppresses
  11. L18:2467 weakly excites
  12. L8:127 weakly suppresses
  13. L20:2825 weakly excites
  14. L14:1853 weakly suppresses
  15. L13:167 weakly excites
  16. L9:1592 weakly excites
  17. L10:1611 weakly suppresses
  18. L14:2664 weakly suppresses
  19. L18:3068 weakly excites
  20. L19:1126 weakly excites
  21. L12:3030 weakly excites
  22. L15:2408 weakly excites
  23. L11:11 weakly suppresses
  24. L16:1121 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 -3.92 and -3.781 values between -2.39 and -2.251 values between -1.70 and -1.561 values between -1.42 and -1.283 values between -1.14 and -1.012 values between -1.01 and -0.872 values between -0.87 and -0.734 values between -0.73 and -0.5910 values between -0.59 and -0.4544 values between -0.45 and -0.31237 values between -0.31 and -0.171019 values between -0.17 and -0.032000 values between -0.03 and 0.10406 values between 0.10 and 0.2493 values between 0.24 and 0.3840 values between 0.38 and 0.5210 values between 0.52 and 0.668 values between 0.66 and 0.809 values between 0.80 and 0.942 values between 0.94 and 1.086 values between 1.08 and 1.214 values between 1.21 and 1.351 values between 1.35 and 1.494 values between 1.49 and 1.63offnegativepositiveoften

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. 滘
  2. estic
  3. 孳
  4. aurus
  5. -parent
  6. ệu
  7. hủy
  8. urrection
  9. нь
  10. ports

Suppresses

  1. schließen
  2. utilus
  3. plugins
  4. 该公司
  5. utschen
  6. ?」
  7. FileInfo
  8. USED
  9. ечно
  10. chili