Neuron Explorer

Qwen3-0.6B

'in' or 'on' before a place or date

A neuron in layer 13 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 reacts to small words like "in" or "on" that come just before a place, a year or a date, such as "published on" or "in Chemistry". When turned up, the model wrote unrelated text, so it does not show a clear effect on writing.

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

    war, France was ravaged by civil strife as its population, economy and resources depleted.⏎- ^ Swyrich, Archive materials⏎This page was last modified on 5 January <|endoftext|>Printable Graph Paper For Room Layout – What’s

  2. #2 peak

    Retrieved from https://youtu.be/qlSxPouPCIM⏎The post 1929 Stock Market Crash and the Great Depression. first appeared on COMPLIANT PAPERS.⏎The post 1929 Stock

  3. #3 peak

    ed by the seaplane hull or float when held at a fixed trim angle. It is considered positive when the bow tends to rise. tunnel, water-See WATER TUNNEL. tunnel, wind-See WIND TUNNEL.

  4. #4 peak

    key role of ubiquitin in basic cellular processes was cemented with the awarding of the 2004 Nobel Prize in Chemistry to ubiquitin research (3).Ubiquitin (Figure 1)

  5. #5 peak

    King Albert to the University of Ghent by cheering ‘Buffalo! Buffalo!’ in 1913.⏎During the 1920 Olympic games in Antwerp, the Belgian athlete Omar Smet, who was a member of the

  6. #6 peak

    awuta, Volyn’, present-day Ukrainian SSR; died Feb. 28, 1866, in Cudnow. Count and Polish writer.⏎In the historical narrative Soplica

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: if the the there the the there there there see see see see see see see see see see see see see see see see see see see see see see see see see see see see see see see see see see see see see see

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

    they also tell us about what it was like when it was young. That’s due to gravitational lensing, a peculiar effect of relativity where the overall gravity of the cluster bends the light of an even more distant galaxy behind it, distorting

  2. #2 peak

    cannot recall the date of their last period, have an irregular menstrual cycle, or who have conceived whilst breastfeeding or soon after stopping the pill. We measure the size of the fetus and from this we calculate the expected date of delivery.⏎- To diagnose

  3. #3 peak

    Drill a 1/4" hole in the flat end of the broom stick.⏎7. Drill two shallow 1/8" holes on the sides of the broomstick, positioned⏎so you can fit U-shaped wire over the

  4. #4 peak

    the filament of the lamp, that we are using through a battery source can be controlled by the rheostat. Now by controlling the incoming current, the brightness of the filament is increased or decreased.⏎Going through this process there will be a certain

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. L4:16 excites
  2. L4:4 suppresses
  3. L10:272 suppresses
  4. L10:155 suppresses
  5. L6:751 excites
  6. L8:636 excites
  7. L8:1003 suppresses "the" or "of" before a noun phrase
  8. L11:153 weakly suppresses
  9. L7:454 weakly excites
  10. L10:469 weakly suppresses 'of' after a noun phrase
  11. L7:751 weakly suppresses
  12. L6:388 weakly excites
  13. L8:45 weakly suppresses
  14. L6:61 weakly excites
  15. L12:1276 weakly suppresses
  16. L6:1534 weakly suppresses
  17. L10:39 weakly excites
  18. L1:25 weakly suppresses
  19. L3:1114 weakly suppresses
  20. L9:651 weakly excites
  21. L3:344 weakly suppresses
  22. L11:1168 weakly suppresses
  23. L5:453 weakly suppresses
  24. L12:2833 weakly excites

Feeds later layers

  1. L14:830 excites
  2. L16:896 excites
  3. L16:1320 suppresses
  4. L14:313 suppresses
  5. L18:606 excites
  6. L14:672 excites
  7. L15:1299 excites
  8. L18:134 suppresses
  9. L14:280 suppresses
  10. L15:202 excites mixed, digits in URLs and comparisons
  11. L16:786 excites
  12. L15:1381 excites
  13. L18:156 excites
  14. L20:375 suppresses
  15. L14:739 excites 'the' or 'your' before a body part or place
  16. L19:1129 excites
  17. L16:906 excites
  18. L18:63 excites
  19. L25:16 suppresses
  20. L18:370 excites
  21. L14:20 suppresses
  22. L21:176 excites
  23. L16:297 excites
  24. L23:24 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.

1 values between -6.75 and -6.481 values between -5.69 and -5.431 values between -4.64 and -4.381 values between -3.85 and -3.582 values between -3.58 and -3.321 values between -3.32 and -3.063 values between -2.79 and -2.531 values between -2.53 and -2.275 values between -2.27 and -2.007 values between -2.00 and -1.749 values between -1.74 and -1.4811 values between -1.48 and -1.2112 values between -1.21 and -0.9531 values between -0.95 and -0.6951 values between -0.69 and -0.42324 values between -0.42 and -0.162397 values between -0.16 and 0.10876 values between 0.10 and 0.37118 values between 0.37 and 0.6325 values between 0.63 and 0.907 values between 0.90 and 1.1612 values between 1.16 and 1.423 values between 1.42 and 1.694 values between 1.69 and 1.953 values between 1.95 and 2.211 values between 2.21 and 2.481 values between 3.53 and 3.79offnegativepositiveoften

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. bing
  2. 辛
  3. inations
  4. 余人
  5. ework
  6. aters
  7. ore
  8. *
  9. anship
  10. ett

Suppresses

  1. @section
  2. 摘编
  3. o
  4. yaptır
  5. обрат
  6. hread
  7. dì
  8. 个百分
  9. umper
  10. -parts