Neuron Explorer

Qwen3-0.6B

capital K starting place or group names

A neuron in layer 25 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 a capital K that starts a proper name, especially place names like Kola, Kaliningrad, Kazakhstan and Kassites. The strong negative side is the words 'team' and 'of' before 'members' or 'elements'. Turned up, the model repeated the word 'member'.

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

    The American Heritage® Dictionary of the English Language, 4th Edition⏎- A region of extreme northern Europe including northern Norway, Sweden, and Finland and the Kola Peninsula of northwest Russia. It is largely within the Arctic Circle.⏎from Wik

  2. #2 peak

    the map was a great way to combine geographical learning and world history.”⏎An important stop in the deserts of present-day Kazakhstan, Otrar on Kaitlin’s map, is now just ruins.⏎Unlike an essay, the map

  3. #3 peak

    1913.⏎During the 1920 Olympic games in Antwerp, the Belgian athlete Omar Smet, who was a member of the KAA Gent athletic department, responded to the rhythmical cheering of the American delegation with

  4. #4 peak

    .⏎In the Congo-Kinshasa Province, the Sisters believe that protection of the environment is an essential

  5. #5 peak

    foods or supplements.<|endoftext|>Bagdannen, a former Mennonite congregation near Tilsit, East Prussia (now Sovetsk, Kalingrad Oblast, Russia), also called "the scattered church of Prussian

  6. #6 peak

    which comprise 31% of the population. Equatorial Bantu make up an estimated 19% of the population. Kirdi and Fulani peoples are also a good percentage of the residents of Cameroon

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: its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its its

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

    tend to be more enthusiastic and will develop a greater sense of camaraderie than within a traditional team. This type of teamwork creates a shared sense of responsibility among team members as well as encouraging a greater sense of loyalty to the organization. Self-directed

  2. #2 peak

    observed when formal leaders experience a heightened subjective sense of power. As a consequence, they will tend to dominate group discussions. This, in turn, will lead other team members to perceive that their perspectives are not valued.⏎Individuals experiencing a heightened sense

  3. #3 peak

    100 fundamental concepts and about 250 heuristics it discovered several important mathematical concepts including subsets, disjoint sets, sets with the same number of elements, and numbers. It worked by filling slots in frames maintaining an agenda of

  4. #4 peak

    chart or the structure of the company and the relationships among its jobs and responsibilities is set up as follows: CEO, Vice President, Director,

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. L22:1117 excites
  2. L21:971 excites
  3. L20:1914 excites
  4. L21:840 excites
  5. L15:2473 excites
  6. L22:1921 excites
  7. L13:1320 suppresses
  8. L19:2469 suppresses
  9. L21:1248 suppresses
  10. L17:2303 weakly suppresses
  11. L19:1909 weakly excites
  12. L14:2916 weakly excites
  13. L18:1522 weakly excites
  14. L11:233 weakly suppresses
  15. L13:2063 weakly excites
  16. L19:2122 weakly suppresses
  17. L19:254 weakly suppresses
  18. L23:3003 weakly excites
  19. L13:976 weakly suppresses
  20. L8:298 weakly excites
  21. L19:2379 weakly suppresses
  22. L7:623 weakly suppresses
  23. L13:2881 weakly suppresses
  24. L4:2621 weakly suppresses chemistry name endings like ether and ketone

Feeds later layers

  1. L27:2478 excites
  2. L27:1047 weakly suppresses
  3. L27:1162 weakly suppresses
  4. L27:1613 weakly excites
  5. L27:1961 weakly excites
  6. L26:1024 weakly excites
  7. L27:1187 weakly excites
  8. L27:1688 weakly excites
  9. L27:1215 weakly excites
  10. L27:1816 weakly excites
  11. L27:1139 weakly excites
  12. L27:1102 weakly suppresses
  13. L27:1703 weakly excites
  14. L27:1426 weakly excites
  15. L26:2284 weakly excites
  16. L27:2095 weakly excites
  17. L26:1469 weakly excites
  18. L27:2155 weakly excites
  19. L27:1397 weakly excites "the" in "to the next"
  20. L27:1737 weakly suppresses
  21. L26:2244 weakly suppresses
  22. L27:330 weakly excites
  23. L26:2869 weakly excites
  24. L26:2136 weakly 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 -25.16 and -24.061 values between -21.87 and -20.781 values between -14.20 and -13.111 values between -13.11 and -12.011 values between -12.01 and -10.921 values between -10.92 and -9.823 values between -9.82 and -8.731 values between -8.73 and -7.633 values between -7.63 and -6.545 values between -6.54 and -5.446 values between -5.44 and -4.3519 values between -4.35 and -3.2539 values between -3.25 and -2.1678 values between -2.16 and -1.062256 values between -1.06 and 0.041274 values between 0.04 and 1.13121 values between 1.13 and 2.2342 values between 2.23 and 3.3228 values between 3.32 and 4.427 values between 4.42 and 5.5111 values between 5.51 and 6.611 values between 6.61 and 7.703 values between 7.70 and 8.801 values between 8.80 and 9.891 values between 10.99 and 12.081 values between 12.08 and 13.182 values between 17.56 and 18.66offnegativepositiveoften

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. members
  2. elements
  3. elements
  4. members
  5. member
  6. Elements
  7. member
  8. -elements
  9. .elements
  10. öğ

Suppresses

  1. its
  2. Its
  3. Its
  4. 它的
  5. 其
  6. iran
  7. ак
  8. 她的
  9. ティブ
  10. eur