Neuron Explorer

Qwen3-0.6B

capital K starting a word

A neuron in layer 27 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 lights up on a capital K that starts a word or name, such as Kilo, KAYAK, Kurd or Kathy. When it was turned up, the model's text started repeating the word 'inventor' over and over instead of making sense.

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

    During the heat of the Battle of Badr, the Prophet Muhammad was said to have done what?⏎Thrown sand into the air in the direction of the enemy⏎Killed those who fled⏎The Battle of Badr is a battle in Islamic history

  2. #2 peak

    and canteen effluent waste. The above waste is treated through Pond setup in the institute with a capacity of 24000 KLD (Kilo Liters per Day). The entire treated water is used for watering the gardens

  3. #3 peak

    report Expanding Ireland’s Marine Protected Area Network can be found here.<|endoftext|>Origin and Etymology of kayak⏎First Known Use: 1757⏎KAYAK Defined for English Language Learners⏎Definition of kayak for English Language Lear

  4. #4 peak

    1.8 million Kurds living in Syria. This number, however, is anything but accurate as the Syrian government does not keep figures. Also the definition of ‘Kurd’ in Syria is open to debate . Regardless of exact numbers, Kurds

  5. #5 peak

    patients.⏎Keyword: Cognitive therapy, depression score⏎Penulis: Wening Marsudi Astuti, Made Sumarwati, Tulus Seyiono⏎Kode Jurnal: jpkeperawatandd10000

  6. #6 peak

    human rights in Tunisia. As the first black woman we have ever seen in the Tunisian Parliament, I am proud to be part of that minority, always.”⏎Ks<|endoftext|>Select one (1) column from a dataset with whole non-negative integers

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: instead of spending time drinking water instead of drinking water drinking water drinking water drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking drinking

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

    ) Thus, seeing the invincible wolves increasing, all the women⏎and men of vraja as a group, said:⏎sthAne neha na nah kAryaM vrajAmo.anyanmahadvanam |⏎y

  2. #2 peak

    tomatoes, tomatillos, camotes (sweet potatoes), jicama (a tuber also known as sengkwang, yam bean, singkamas, Mexican turnip), amaranth (also called pigweeds)

  3. #3 peak

    is accepted which rules out the other three. The alternatives are:⏎- Does one karma produce one birth?⏎- Does one karma cause several births?⏎- Do several karmas result in several births?⏎- Do several karmas give rise to

  4. #4 peak

    ĪW, a Middle Iranian word (in Manichean Mid. Pers. script gryw, rarely gryyw; in Sogd. script kryw; in Pahl. script rendered by the ideogram, q.v

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. L20:613 strongly suppresses
  2. L22:2170 strongly excites
  3. L18:816 suppresses
  4. L21:590 excites
  5. L19:2402 excites
  6. L10:90 excites
  7. L17:1216 suppresses
  8. L1:53 suppresses
  9. L18:131 excites
  10. L17:437 excites
  11. L19:381 excites word-start 'acc' fragments
  12. L16:47 suppresses
  13. L0:1462 suppresses
  14. L18:288 excites
  15. L16:184 suppresses
  16. L0:1878 excites
  17. L16:67 suppresses
  18. L20:97 weakly excites
  19. L1:50 weakly excites
  20. L19:243 weakly excites
  21. L2:1859 weakly suppresses
  22. L3:1116 weakly suppresses
  23. L20:371 weakly suppresses
  24. L24:2013 weakly excites

Feeds later layers

None: this is the last layer.

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 -18.95 and -17.031 values between -13.19 and -11.272 values between -3.59 and -1.673829 values between -1.67 and 0.2565 values between 0.25 and 2.184 values between 2.18 and 4.102 values between 4.10 and 6.021 values between 6.02 and 7.942 values between 11.78 and 13.701 values between 55.95 and 57.88offnegativepositiveoften

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. wargs
  2. inventor
  3. raries
  4. Merit
  5. iller
  6. ont
  7. 顯
  8. Hz
  9. automatically
  10. 实质

Suppresses

  1. ึง
  2. copies
  3. ща
  4. x
  5. .getBounds
  6. ballo
  7. utations
  8. 废弃物
  9. rematch
  10. remain