Neuron Explorer

Qwen3-0.6B

second half of proper names

A neuron in layer 6 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the end part of a name, especially names of people, places, courts and organisations in history and news, such as 'Board' in Brown v. Board. Its negative side fires on the end of journal names like 'Physician' or 'Bulletin'. Turning it up made the text wander without a clear theme.

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

    have little understanding of what determines variation among individuals for voluntary exercise levels," Garland said.⏎Study results appear online Sept. 1 in the Proceedings of the Royal Society B.⏎The researchers began their experiments in 1993 with 22

  2. #2 peak

    une Function, 2017, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5707683/#:~:text=It%20is%20a%20potent%

  3. #3 peak

    are unfair … to Tammany Hall. The corrupt Irish politicians who kept that famously well-oiled patronage machine of

  4. #4 peak

    such as Countee Cullen, Arna Bontemps, author of Drums at Dusk, and writer and folklorist Zora Neale Hurston.⏎Words and Deeds in American History: Selected Documents Celebrating the

  5. #5 peak

    Organization has expressed concern over an upsurge in COVID-19 mortality following a 10-week drop.⏎The WHO Director-General, Tedros Adhanom, revealed this at a COVID-19 media event on Monday in Geneva.⏎

  6. #6 peak

    the two crusading lawyers who led the legal battle to end segregation, one case and one courtroom at a time.⏎The Supreme Court’s decision inBrown v. Board of Educationis widely considered a seminal point in the battle to end segregation, but

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: I need to get this done. I need to get this done. I need to get this done. I need to get this done. I need to get this done. I need to get this done. I need to get this done

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

    you are going to say. However, don’t make the presentation too word<|endoftext|>Sore Throat Complications Not Increased with Fewer Antibiotics⏎Am Fam Physician. 2006 May 1;73(9):1

  2. #2 peak

    Vintage, 1996.⏎Silver, Larry. “Nature and Nature’s God: Landscape and Cosmos of Albrecht Altdorfer.” Art Bulletin 81.2 (June 1999): 19

  3. #3 peak

    recent study published in the journal Aggressive Behavior supports the idea that perfectionism may develop in a social group and suggests that indirect

  4. #4 peak

    the world's greatest feats of construction by an archaic civilization of hunters and gatherers.⏎Now, new research in the current issue of the journal Geoarchaeology, offers compelling evidence that one of the massive earthen mounds at Poverty Point

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. L5:2917 excites
  2. L5:1439 suppresses
  3. L5:11 excites
  4. L5:67 excites
  5. L4:60 suppresses
  6. L4:84 suppresses
  7. L5:315 suppresses
  8. L3:256 excites
  9. L5:14 suppresses
  10. L3:521 suppresses
  11. L5:1081 excites
  12. L5:780 suppresses
  13. L5:76 suppresses
  14. L3:607 excites
  15. L3:1124 suppresses
  16. L1:678 excites
  17. L5:998 suppresses
  18. L4:133 excites
  19. L5:666 excites
  20. L0:1173 suppresses
  21. L2:887 excites
  22. L2:487 suppresses
  23. L3:323 excites
  24. L5:2135 suppresses

Feeds later layers

  1. L7:22 strongly excites
  2. L7:784 suppresses
  3. L8:1 excites
  4. L16:182 excites
  5. L18:51 excites
  6. L23:3 excites
  7. L14:21 excites
  8. L12:2 excites
  9. L7:214 suppresses
  10. L25:5 excites
  11. L14:12 excites
  12. L11:41 excites
  13. L16:166 excites
  14. L14:1 excites
  15. L10:0 excites
  16. L16:6 excites
  17. L12:2740 suppresses
  18. L18:18 excites
  19. L16:201 excites
  20. L14:6 excites
  21. L18:25 excites
  22. L16:35 excites 'not' in 'but not limited to'
  23. L18:38 excites
  24. L18:9 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.

2 values between -4.40 and -4.181 values between -2.65 and -2.431 values between -1.99 and -1.773 values between -1.55 and -1.331 values between -1.33 and -1.115 values between -1.11 and -0.897 values between -0.89 and -0.6715 values between -0.67 and -0.4537 values between -0.45 and -0.2472 values between -0.24 and -0.021231 values between -0.02 and 0.201956 values between 0.20 and 0.42551 values between 0.42 and 0.6419 values between 0.64 and 0.862 values between 0.86 and 1.081 values between 1.30 and 1.521 values between 1.74 and 1.961 values between 2.40 and 2.621 values between 2.84 and 3.061 values between 4.15 and 4.37offnegativepositiveoften

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. ful
  2. prior
  3. ranges
  4. metric
  5. data
  6. hot
  7. gate
  8. range
  9. 的话
  10. free

Suppresses

  1. }s
  2. slaught
  3. ”)
  4. íveis
  5. updatedAt
  6. ”?
  7. avis
  8. ⽰
  9. chers
  10. ”.⏎⏎