Neuron Explorer

Qwen3-0.6B

end of web page disclaimer

A neuron in layer 18 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 is strongest at the invisible end-of-page marker that comes after a medical or health disclaimer, such as 'consult your doctor'. Its negative side is weak and mostly about numbers. Steering did not produce a clear health theme, just rambling or repeated text.

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

    direct supervision of a qualified health practitioner such as a naturopathic physician.<|endoftext|>In 1939 the Harvard sociologist Edward Hartshorne gathered eyewitness

  2. #2 peak

    intended to be a substitute for professional medical advice, diagnosis, or treatment. Always seek the advice of qualified health providers with questions you may have regarding medical conditions.<|endoftext|>The Amendment Process and the African American Fight for Representation and Equality⏎Black History Month

  3. #3 peak

    given as suggestions only. We recommend you also consult your healthcare provider, and urge you to contact them immediately if your question is urgent or about a medical condition.<|endoftext|>The Joint United Nations Programme on HIV/AIDS (UNAIDS) confirmed that in

  4. #4 peak

    treat patients who only have inflammation in the stomach. Triple therapy is not right for every child. Ask your doctor if it is the right treatment for your child.<|endoftext|>Driving an embedded car save not just time but it can save lives as well.

  5. #5 peak

    , each of which is shown in turn below:⏎The distance between any grid point and the rectangular border is one unit of measurement (equal to the length of the<|endoftext|>Shree Ganeshay Namah⏎How to write Anchoring Script in English

  6. #6 peak

    ions in the region, and are insufficient to offset present rates of relative sea level rise. © 2006 Elsevier Ltd. All rights reserved.<|endoftext|>Are you tired of being sick and tired of being sick and tired? Then,

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: it's a more interesting and more interesting than the materialistic and materialistic stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff stuff

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

    13, 81 ce), Roman emperor (79–81), and the conqueror

  2. #2 peak

    Egypt, El Salvador, Finland, Grenada, Guatemala, Japan, Lao, PDR, Macedonia,

  3. #3 peak

    1, 1, 2, 3, 5, 8, 1

  4. #4 peak

    1, 1/4 turn clockwise and then back 2_______.⏎5.1

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. L17:2635 strongly excites
  2. L17:2002 strongly excites
  3. L15:1898 excites
  4. L17:1527 excites
  5. L17:1586 excites
  6. L13:257 excites
  7. L16:1430 suppresses
  8. L17:2327 suppresses
  9. L16:562 suppresses
  10. L14:978 suppresses
  11. L16:2077 excites
  12. L13:1860 excites
  13. L16:2962 suppresses unclear: weak mid-phrase prepositions
  14. L17:2895 excites
  15. L13:2182 suppresses
  16. L15:2759 suppresses
  17. L14:2525 suppresses
  18. L16:2943 excites
  19. L17:1635 excites
  20. L14:66 suppresses
  21. L15:158 suppresses copula after 'probably' or 'may'
  22. L15:1959 excites
  23. L16:1313 excites
  24. L16:631 excites

Feeds later layers

  1. L19:1567 strongly excites
  2. L24:174 excites
  3. L20:759 excites
  4. L19:355 suppresses
  5. L20:697 excites
  6. L19:1673 excites
  7. L19:401 suppresses
  8. L21:295 suppresses
  9. L24:367 excites
  10. L23:348 excites
  11. L19:823 suppresses
  12. L19:2218 excites
  13. L19:1336 excites
  14. L20:708 suppresses
  15. L19:1291 excites
  16. L21:457 excites
  17. L19:225 suppresses
  18. L19:1446 excites
  19. L22:865 excites
  20. L19:857 excites
  21. L20:1039 suppresses
  22. L21:403 excites
  23. L19:1789 excites middle of personal names
  24. L27:141 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 -5.67 and -5.141 values between -5.14 and -4.622 values between -4.62 and -4.096 values between -4.09 and -3.5610 values between -3.56 and -3.0313 values between -3.03 and -2.5127 values between -2.51 and -1.9855 values between -1.98 and -1.4594 values between -1.45 and -0.92324 values between -0.92 and -0.402564 values between -0.40 and 0.13589 values between 0.13 and 0.66122 values between 0.66 and 1.1947 values between 1.19 and 1.7229 values between 1.72 and 2.2410 values between 2.24 and 2.774 values between 2.77 and 3.301 values between 3.30 and 3.831 values between 3.83 and 4.361 values between 4.36 and 4.881 values between 5.41 and 5.941 values between 7.52 and 8.051 values between 9.11 and 9.631 values between 9.63 and 10.161 values between 11.22 and 11.741 values between 14.38 and 14.911 values between 14.91 and 15.44offnegativepositiveoften

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. night
  2. shore
  3. night
  4. edy
  5. -night
  6. 夜
  7. Crypt
  8. otope
  9. Night
  10. Nights

Suppresses

  1. ')")⏎
  2. '))⏎
  3. ĩnh
  4. >*/⏎
  5. %@
  6. chal
  7. 眷
  8. +%
  9. 詳
  10. ()}}⏎