Neuron Explorer

Qwen3-0.6B

em dash opening an aside

A neuron in layer 23 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 long dash (, ) placed in the middle of a sentence to set off an extra phrase. Turning it up made the model repeat phrases like 'the same day', which does not clearly relate to the dash.

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

    than half of the<|endoftext|>Shin splints” is a term for a spectrum of pain in the front of the shin. The condition—very common in runners—can result from a stress injury (irritation of the bone) or a

  2. #2 peak

    ). During the French Revolution (1789–99), the convent—continuously occupied and much expanded since its foundation in the thirteenth century—was closed and the nuns expelled. The convent was then sold and many of

  3. #3 peak

    series that demonstrate both the scale and specific locations of coastal change. Documentation of current delta changes—including coastal development along the Yellow River several other rapidly changing coastlines— continues from the International Space Station.⏎You can also see how the Yellow River Delta

  4. #4 peak

    home if left untreated. A dehumidifier or air conditioner can help.⏎Answer: True. Radon—a colorless, tasteless and odorless gas—seeps into homes through cracks in walls, floors and foundations. If radon

  5. #5 peak

    an observatory that can track this for years,” he said. “This allows us to use new technologies to study the how the ecology—the role of plants specifically—relates to the hydrology, the movement of water.<|endoftext|>Details of a

  6. #6 peak

    00 BCE.⏎“The occurrence of these two virulent pathogens at the end of the Early Minoan period in Crete,” they wrote in their paper, “emphasizes the necessity to re-introduce infectious diseases as an additional factor possibly

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: what if you have an animal that can be just like a person? what if you could use it to study and treat different sciences, but it would be a different and even better way to study and treat by just learning from it? to

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

    erary cult that were no longer current at the time the pyramids were built. This indicates that at least some of the Pyramid Texts can be dated to Pre-Dynastic times.⏎The texts may provide clues to the thinking behind the development

  2. #2 peak

    decreased blood⏎SETTING: Homes of participants.⏎PARTICIPANTS: 35 volunteers⏎with NIDD will be recruited by advertising on television and radio.⏎INTERVENTIONS: All participants⏎will receive education on dietary supplemention with Jerusalem art

  3. #3 peak

    8 of 101⏎It looks like you're using a mobile device. We recommend using a physical keyboard for transcription entry.⏎28 REPORT NATIONAL ADVISORY COMMITTEE FOR AERONAUTICS transverse-continued. main trans

  4. #4 peak

    account of race or color.⏎With the Civil Rights Act turning 50, the success of the struggle for equal rights for women and minorities<|endoftext|>The No Child Left Behind Act, signed into law by President Bush January 8, 20

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. L20:563 strongly excites
  2. L21:418 strongly excites
  3. L21:959 strongly suppresses
  4. L19:1240 strongly suppresses
  5. L21:846 strongly excites
  6. L21:118 strongly suppresses
  7. L19:124 strongly suppresses
  8. L20:473 suppresses
  9. L22:65 suppresses
  10. L22:54 suppresses
  11. L20:296 suppresses
  12. L19:484 excites
  13. L21:533 excites
  14. L20:605 suppresses
  15. L20:59 suppresses
  16. L20:1013 excites
  17. L22:111 excites
  18. L19:68 excites full stops closing a claim
  19. L16:1421 suppresses
  20. L19:305 suppresses
  21. L19:2730 suppresses
  22. L18:91 excites
  23. L19:1174 excites
  24. L19:96 excites

Feeds later layers

  1. L25:557 strongly excites
  2. L24:328 excites
  3. L24:650 excites
  4. L26:186 excites
  5. L25:595 excites
  6. L24:483 excites
  7. L27:337 excites
  8. L26:357 excites
  9. L26:352 excites
  10. L26:553 excites
  11. L26:1547 excites
  12. L27:811 excites
  13. L26:788 excites
  14. L26:418 excites
  15. L27:192 excites
  16. L26:1426 excites
  17. L26:389 excites
  18. L26:503 excites
  19. L26:696 excites
  20. L27:561 excites
  21. L27:732 excites
  22. L27:1194 excites
  23. L26:465 excites
  24. L27:945 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.

3 values between -11.92 and -10.514 values between -9.10 and -7.696 values between -7.69 and -6.2813 values between -6.28 and -4.879 values between -4.87 and -3.45405 values between -3.45 and -2.042438 values between -2.04 and -0.63649 values between -0.63 and 0.78112 values between 0.78 and 2.1971 values between 2.19 and 3.6045 values between 3.60 and 5.0149 values between 5.01 and 6.4322 values between 6.43 and 7.8415 values between 7.84 and 9.2514 values between 9.25 and 10.6616 values between 10.66 and 12.074 values between 12.07 and 13.484 values between 13.48 and 14.899 values between 14.89 and 16.314 values between 16.31 and 17.722 values between 17.72 and 19.132 values between 19.13 and 20.541 values between 20.54 and 21.953 values between 23.36 and 24.771 values between 27.60 and 29.011 values between 29.01 and 30.422 values between 31.83 and 33.241 values between 33.24 and 34.651 values between 40.30 and 41.711 values between 41.71 and 43.121 values between 43.12 and 44.53offnegativepositiveoften

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. ass
  2. ens
  3. im
  4. off
  5. (st
  6. (s
  7. in
  8. ongo
  9. en
  10. ad

Suppresses

  1. 薮
  2. Rgb
  3. -Jun
  4. uent
  5. ạm
  6. Afro
  7. département
  8. lesbian
  9. brisk
  10. libre