Neuron Explorer

Qwen3-0.6B

"of" after a noun of quantity or container

A neuron in layer 12 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 follows a noun that names an amount or container, like "a layer of" or "a period of". Turning it up made the model's text drift into rambling, repeated phrasing rather than a clear topic.

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

    other North American wild mammal. The armadillo’s body is covered with an armored carapace or shell. The carapace is a double layer of horn and bone, segmented into three main divisions: an anterior scapular shield

  2. #2 peak

    er to cool down,” he said.⏎“Just a few months ago, wetlands were brimming with water, but now many are just parched expanses of cracked mud.⏎“Birds need water to drink every day, and when the

  3. #3 peak

    , charcoal or soot-based solid ink, on silk or paper. An infinite range of subtle tones can be used, sometimes with the addition of transparent washes of light colour. The quality and strength of the line depends on the artist's control

  4. #4 peak

    Arthritis is an autoimmune disorder in which your body mistakenly begins attacking the synovium as though they were an invading illness. The synovium is a lining of membranes that surround your joints.⏎When your body attacks this lining, inflammation occurs and

  5. #5 peak

    .⏎Tools and Resources for App Coding⏎To facilitate the process of coding for<|endoftext|>A sprayed seal is a thin layer of binder sprayed onto a pavement surface with a layer of aggregate incorporated and which is imp

  6. #6 peak

    for further auroral displays tonight or tomorrow night if we are lucky.⏎The Northern Lights are a result of a geomagnetic storm. These storms are short-lived periods of high geomagnetic activity where the Earth’s magnetic field

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: I have been told by the students, and the teacher, and the parents and the students, and the teachers, and the scientists, and the students, and the teacher, and the students, and the ... and ... I have something.

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

    When a person dies, the bacteria can move to other parts of the body, which causes the reaction which starts decomposition.⏎The embalming process drains the body of all the liquids that are no longer moving. The formal<|endoftext|>Algebraic geometry

  2. #2 peak

    cheadlines⏎Google has started a new page on Google Plus to share their vision of what its augmented reality glasses could be. They are soliciting suggestions from users on what they would like to see from Project Glass.⏎“We think

  3. #3 peak

    look consistent<|endoftext|>Whether it’s a child or a teenager, a lot of us have a backpack to use when going to school. These backpacks are used to store things such as books, notebooks, and

  4. #4 peak

    Minneapolis. This is a rare technique of covering and insulating silos. The other silos were added in the 1940s and used to store barley. The factory outputted 250,000 (1

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. L6:3070 excites
  2. L10:886 excites
  3. L10:1395 excites
  4. L11:1045 suppresses
  5. L6:2862 suppresses
  6. L9:1025 suppresses
  7. L11:361 suppresses
  8. L9:2219 excites
  9. L6:1956 excites "for" introducing a purpose or source
  10. L8:508 suppresses
  11. L11:1811 suppresses
  12. L11:153 suppresses
  13. L10:2138 weakly excites
  14. L6:285 weakly suppresses
  15. L5:1531 weakly excites
  16. L11:579 weakly suppresses
  17. L11:2303 weakly excites
  18. L6:2099 weakly excites
  19. L6:1666 weakly suppresses
  20. L4:1271 weakly excites
  21. L6:698 weakly suppresses
  22. L6:1693 weakly suppresses
  23. L9:2764 weakly suppresses
  24. L11:3048 weakly suppresses

Feeds later layers

  1. L17:2405 suppresses
  2. L20:1498 suppresses
  3. L17:1574 excites
  4. L16:2852 excites 'in' after a noun or participle
  5. L14:539 suppresses
  6. L13:2136 suppresses
  7. L22:1228 suppresses
  8. L14:1393 suppresses
  9. L14:1192 excites
  10. L17:1164 excites
  11. L14:1728 excites "of" after a numbered section or title
  12. L24:490 suppresses
  13. L17:505 suppresses
  14. L19:1517 suppresses
  15. L16:1948 suppresses
  16. L16:2892 excites
  17. L16:1147 suppresses
  18. L13:1781 suppresses
  19. L19:1277 excites
  20. L15:871 excites
  21. L13:2238 suppresses
  22. L13:1659 suppresses
  23. L18:1635 suppresses
  24. L16:680 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 -2.52 and -2.402 values between -2.15 and -2.031 values between -1.78 and -1.661 values between -1.54 and -1.413 values between -1.41 and -1.292 values between -1.17 and -1.054 values between -1.05 and -0.928 values between -0.92 and -0.804 values between -0.80 and -0.6816 values between -0.68 and -0.5516 values between -0.55 and -0.4372 values between -0.43 and -0.31210 values between -0.31 and -0.18777 values between -0.18 and -0.061942 values between -0.06 and 0.07546 values between 0.07 and 0.19171 values between 0.19 and 0.3163 values between 0.31 and 0.4335 values between 0.43 and 0.5615 values between 0.56 and 0.684 values between 0.68 and 0.804 values between 0.80 and 0.933 values between 0.93 and 1.052 values between 1.05 and 1.173 values between 1.17 and 1.301 values between 1.30 and 1.421 values between 1.67 and 1.791 values between 2.28 and 2.41offnegativepositiveoften

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. ighted
  2. etable
  3. ボード
  4. adesh
  5. etCode
  6. [param
  7. ished
  8. rectangle
  9. lined
  10. ewe

Suppresses

  1. ream
  2. ôle
  3. 람
  4. ")[
  5. utorial
  6. ража
  7. amble
  8. 万物
  9. 潏
  10. 同类