Neuron Explorer

Qwen3-0.6B

'that' starting a clause after a noun

A neuron in layer 16 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 the word 'that' when it introduces a statement after a noun, such as 'the notion that', 'the maxim that' or 'a dream that'. Turning it up made the model get stuck repeating words, so the steering output is not very informative.

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

    ’ birthdays are on the other hand one of the grandest holidays of the year. All this is not surprising given North Korea’s reputation. There is a fact that may be more of an eyebrow raiser.⏎New Years is widely accepted and celebrated

  2. #2 peak

    the system, such as it was, took care of the people (at least some of them).⏎Eventually, Golway argues, Tammany Hall introduced the notion that the government had a social welfare role to play. Laissez-faire capitalism

  3. #3 peak

    Legislature in regards that might be put forth that would direct all the school districts in the state as to how they should address bullying

  4. #4 peak

    Yom Ha' Shoah is not just a Jewish event — it touches everyone,” she says.⏎“Yom Ha' Shoah is a wake-up call that genocide is happening now and we have the power to stop it.”⏎The event is

  5. #5 peak

    at Austin): There's no principle built into the laws of nature that says theoretical physicists have to be happy.⏎NARRATOR: It's a game-changing perspective that opens up a whole new world of possibilities. Coming up: what if new univers

  6. #6 peak

    what comes. These creatively structured poems explore what it means to cultivate respect for self and others, the meaning of good qualities, and to hold fast to the maxim that ‘anything is possible.’⏎Because Brave. A Book about Courage is authored by

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: and to the point: and to the point of notice, and to the point of notice to the point of notice that to the point of being, but, then, to this, to the point of being) etc. etc. and

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

    .⏎On the other side of the world, the Mayans are believed to have utilized the gum from the manilkara chicle evergreen tree (yes, that’s the inspiration for “Chiclet”) in a

  2. #2 peak

    samples incorrectly, an oil-analysis program is a waste of time and money. Key guidelines include:⏎- Take warmed-up machine samples while equipment is running. If that’s not possible, sample no later than 30 minutes after shutdown⏎-

  3. #3 peak

    form of life, they point out, this is simply an example of a normal form of earth life adapting in a remarkable way to harsh conditions. Well, if that's the case, I don't know what all the fuss is about. We

  4. #4 peak

    for New Scientist and estimated that PCB levels would only rise above normal ocean concentration within 1 kilometre of the ships and are unlikely to harm marine life even in that area. "It's not an ethical thing to do," Lohmann says

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. L4:698 excites
  2. L3:1003 excites
  3. L6:1105 weakly excites 'that' after a finding or surprise verb
  4. L4:227 weakly suppresses
  5. L0:2204 weakly excites
  6. L0:1216 weakly excites
  7. L9:661 weakly excites
  8. L10:33 weakly suppresses
  9. L6:558 weakly suppresses
  10. L15:181 weakly excites
  11. L3:2786 weakly suppresses
  12. L8:132 weakly suppresses
  13. L1:2278 weakly excites
  14. L3:2141 weakly excites
  15. L13:112 weakly suppresses
  16. L3:1253 weakly suppresses
  17. L11:73 weakly excites
  18. L8:369 weakly suppresses 'of' after 'learned' or 'because'
  19. L15:397 weakly suppresses
  20. L5:1061 weakly excites
  21. L1:945 weakly suppresses
  22. L12:728 weakly suppresses
  23. L1:471 weakly suppresses
  24. L11:1774 weakly excites

Feeds later layers

  1. L17:2695 excites
  2. L17:1361 excites 'that' after verbs like recommend
  3. L23:298 excites
  4. L17:2280 excites
  5. L17:1282 excites
  6. L19:1100 excites
  7. L21:191 suppresses
  8. L19:2919 excites
  9. L17:2684 excites
  10. L21:639 excites
  11. L20:982 suppresses
  12. L23:147 excites "that" introducing a clause or "that is"
  13. L19:94 suppresses
  14. L18:1519 excites
  15. L19:433 excites
  16. L19:316 weakly suppresses
  17. L19:695 weakly suppresses
  18. L17:2363 weakly suppresses
  19. L22:892 weakly excites
  20. L18:229 weakly suppresses
  21. L22:360 weakly excites
  22. L21:721 weakly excites
  23. L26:2986 weakly suppresses
  24. L18:598 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.

53 values between -1.17 and -0.80892 values between -0.80 and -0.432286 values between -0.43 and -0.05561 values between -0.05 and 0.3236 values between 0.32 and 0.6917 values between 0.69 and 1.0711 values between 1.07 and 1.443 values between 1.44 and 1.824 values between 1.82 and 2.193 values between 2.19 and 2.565 values between 2.56 and 2.943 values between 2.94 and 3.311 values between 3.31 and 3.681 values between 3.68 and 4.063 values between 4.06 and 4.431 values between 4.43 and 4.802 values between 4.80 and 5.182 values between 5.18 and 5.552 values between 5.92 and 6.301 values between 6.30 and 6.671 values between 6.67 and 7.041 values between 7.79 and 8.162 values between 9.28 and 9.661 values between 9.66 and 10.031 values between 10.03 and 10.403 values between 10.40 and 10.783 values between 11.15 and 11.533 values between 11.53 and 11.901 values between 11.90 and 12.272 values between 12.27 and 12.651 values between 12.65 and 13.021 values between 13.02 and 13.391 values between 13.39 and 13.77offnegativepositiveoften

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. 一旦
  2. 一切都是
  3. tero
  4. _"
  5. 由于
  6. 我们知道
  7. "***
  8. {}".
  9. if
  10. quete

Suppresses

  1. gress
  2. ��
  3. iking
  4. zelf
  5. oses
  6. ough
  7. ours
  8. Falls
  9. idine
  10. asions