Neuron Explorer

Qwen3-0.6B

the word 'two' after 'the other'

A neuron in layer 14 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 the word 'two' when the text is talking about a pair or the second of a set, such as 'the other two ships' or 'these two functions'. It is a counting-of-pairs detector. Turning it up did not produce any clear link to pairs, though, so the steering output does not confirm the idea.

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

    one side of an equilateral triangle and the circle touches the other two sides of the triangle. A smaller circle touches the larger circle and. . .

  2. #2 peak

    species of bed bugs worldwide, but luckily for us, only four of them feed on human blood. Of those four, only two prefer human blood. The other two species will feed on humans if their other prey is unavailable, but these infest

  3. #3 peak

    to 14.7 meters), and they were between 8 and 16 feet (2.5 and 5 m) wide.⏎The other two ships the researchers examined were galleys — long, oared ships, about

  4. #4 peak

    require two functions in the Arduino program or sketch, which are setup () and a loop (). The rest of the functions have to be created outside the bracket with these two functions. The best syntax used in defining function is:⏎- Return type. This

  5. #5 peak

    them are three priest wearing the leopard skin (a sign of ancient priesthood). One of them burns incense and presents offerings of food and drink,⏎- The other two are about to perform the important

  6. #6 peak

    is what we call a founder: she came from the Philippines, and her genes were not represented in the North American population. She is also the mom to the two piglets running around the exhibit. Alibangbang and Hinigugma

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: if I feel, if I think, if I feel, if I feel, if I feel, I could have been a Different, If I Could Have Been, I Could Have Been, If I Could Have, Maybe, I Could Have

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

    of cell until the conditions are good for a new sponge to develop. Then the archaeocyte "hatches" and develops into a new mature sponge. The other kind of reproduction that s

  2. #2 peak

    , British Columbia, could bring only good to that area. Many island residents disagree with that conclusion. The distance between the two areas may make any proposal for a bridge cost prohibitive…⏎Extract of sample⏎There

  3. #3 peak

    legal Chinese government.⏎Since embarking on landmark reforms beginning in the late 1970s, the country has emerged from isolation to become the world’s second-largest economy.⏎Xi has made national renewal a central theme of his administration, wanting

  4. #4 peak

    know the things that should be considered when having a guinea pig at home.⏎As rodents, guinea pigs are considered large weighing approximately 700 and 1200 grams or 1.5 to 2.5 pounds

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. L10:93 strongly suppresses
  2. L11:809 strongly excites
  3. L9:275 strongly suppresses
  4. L10:1722 strongly excites
  5. L10:215 strongly excites
  6. L12:1243 suppresses
  7. L10:1085 excites
  8. L7:415 excites
  9. L13:697 excites
  10. L11:467 suppresses
  11. L5:1109 suppresses
  12. L11:231 excites
  13. L7:284 excites
  14. L11:2101 excites
  15. L13:2603 excites
  16. L10:1901 suppresses
  17. L10:1920 excites second number in a pair
  18. L12:1590 excites
  19. L13:2006 excites
  20. L13:1783 suppresses
  21. L11:628 suppresses
  22. L13:1469 suppresses
  23. L12:1385 suppresses
  24. L8:336 excites

Feeds later layers

  1. L16:744 strongly suppresses
  2. L20:534 strongly excites
  3. L19:2186 strongly suppresses
  4. L17:1365 strongly excites
  5. L15:1097 strongly excites
  6. L24:890 strongly excites
  7. L23:1178 suppresses
  8. L20:686 suppresses
  9. L15:1708 excites
  10. L25:2674 excites
  11. L15:1740 suppresses
  12. L17:2127 excites
  13. L27:1226 excites
  14. L21:2789 suppresses
  15. L19:916 excites
  16. L16:1955 suppresses
  17. L16:3009 excites
  18. L23:358 suppresses
  19. L20:3035 suppresses
  20. L22:816 excites
  21. L16:1127 excites opening bracket after an imperial measurement
  22. L19:1161 excites
  23. L16:2403 suppresses
  24. L24:2090 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 -3.64 and -3.401 values between -2.44 and -2.201 values between -2.20 and -1.963 values between -1.96 and -1.722 values between -1.72 and -1.492 values between -1.49 and -1.253 values between -1.25 and -1.0120 values between -1.01 and -0.7747 values between -0.77 and -0.53191 values between -0.53 and -0.291050 values between -0.29 and -0.052352 values between -0.05 and 0.19156 values between 0.19 and 0.4327 values between 0.43 and 0.666 values between 0.66 and 0.909 values between 0.90 and 1.146 values between 1.14 and 1.387 values between 1.38 and 1.627 values between 1.62 and 1.863 values between 1.86 and 2.103 values between 2.10 and 2.341 values between 2.34 and 2.583 values between 2.58 and 2.811 values between 2.81 and 3.051 values between 3.05 and 3.292 values between 3.29 and 3.531 values between 3.53 and 3.771 values between 4.25 and 4.491 values between 5.68 and 5.92offnegativepositiveoften

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. 双方
  2. 两者
  3. Both
  4. both
  5. 二者
  6. 两国
  7. 这两
  8. Both
  9. 两人
  10. 这两个

Suppresses

  1. ABCDE
  2. 九大
  3. ABCDEFG
  4. 五个
  5. 这几个
  6. ?,?,?,?,
  7. 这三个
  8. 依次
  9. 几家
  10. 的各种