Neuron Explorer

Qwen3-0.6B

commas between author names in citations

A neuron in layer 2 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 comma that comes right after an author's surname in a reference list, such as the comma after 'Chung' in a citation. When it was turned up, the written text drifted into unrelated rambling and did not show 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

    Ho, C. S. H., Lo, C.-M., Chan, K. S.-C., Chan, D. W.-O., & Chung, K. K.-H. (2014, July). Effectiveness

  2. #2 peak

    J.M.; Mortimer, S.R.; Brereton, T.; Redhead, J.W.; Thomas, J.A.; Pywell, R.F.. 2012 Identifying time lags in the restoration

  3. #3 peak

    our fishing communities,<|endoftext|>Suicide and Social Change in China⏎- Cite this article as:⏎- Phillips, M.R., Liu, H. & Zhang, Y. Cult Med Psychiatry (1999) 23:

  4. #4 peak

    This handout provides an example of a revised “show-me” sentence and six exercises for students to practice.|⏎Baines, L., & Kunkel, A. (2010). Going bohemian

  5. #5 peak

    results⏎Authors: Walker, Robert J. (John), 1801-1869

  6. #6 peak

    justify your arguments.⏎references: Gwartney, J. D., Stroup, R. L., Sobel, R. S., & Macpherson, D. A. (2018). Economics: Private⏎and public

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: 0.000000000000000000000000000000000000000000000

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

    pysyväislinkit|⏎|Tila||Julkaistu - 1 syysk. 2021|⏎|OKM-julkaisutyyppi||A1 Julkaistu artikkeli,

  2. #2 peak

    helped spearhead efforts to create a warning system to keep jet aircraft from flying into the sort of engine-seizing ash cast up by Eyjafjallajökull, the low-lying volcano whose plume all but shut down

  3. #3 peak

    astronauts and equipment, according to Reuters.⏎China's Shenzhou spacecraft, based on Russia's Soyuz, has the capacity to ferry up to

  4. #4 peak

    011. Successful docking with an unmanned spacecraft had already been conducted in 2011 itself. But this time, the three astronauts in Shenzhou 9 established that China could precisely maneuver a space capsule to rendezvous with and

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. L1:2580 suppresses
  2. L0:178 weakly suppresses
  3. L0:1301 weakly suppresses
  4. L1:2236 weakly excites
  5. L1:2683 weakly excites
  6. L0:1224 weakly suppresses
  7. L0:1042 weakly suppresses "This" starting a labelled sentence
  8. L1:1732 weakly excites
  9. L1:1464 weakly excites
  10. L1:2197 weakly suppresses
  11. L1:628 weakly suppresses
  12. L1:1857 weakly suppresses
  13. L0:313 weakly suppresses "and" joining paired nouns
  14. L1:844 weakly suppresses
  15. L1:1134 weakly suppresses
  16. L1:2796 weakly excites
  17. L0:802 weakly excites
  18. L0:571 weakly excites
  19. L1:1907 weakly suppresses
  20. L1:1692 weakly excites
  21. L0:3015 weakly excites
  22. L1:1735 weakly excites
  23. L0:191 weakly suppresses
  24. L1:987 weakly excites

Feeds later layers

  1. L5:1027 excites
  2. L16:826 weakly suppresses
  3. L10:2091 weakly suppresses
  4. L7:1945 weakly excites
  5. L16:845 weakly excites
  6. L12:495 weakly excites
  7. L17:1665 weakly suppresses
  8. L23:2123 weakly suppresses
  9. L15:1114 weakly suppresses
  10. L4:560 weakly suppresses
  11. L5:1586 weakly excites
  12. L5:1938 weakly excites
  13. L16:1194 weakly suppresses
  14. L10:612 weakly suppresses full stop ending boilerplate or a title
  15. L13:640 weakly suppresses
  16. L14:2440 weakly excites
  17. L6:770 weakly suppresses
  18. L14:1653 weakly excites
  19. L13:2798 weakly excites
  20. L3:2833 weakly suppresses
  21. L6:1397 weakly suppresses
  22. L4:1713 weakly suppresses
  23. L17:1111 weakly suppresses
  24. L12:2776 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.

1 values between -0.95 and -0.812 values between -0.68 and -0.552 values between -0.55 and -0.4213 values between -0.42 and -0.29158 values between -0.29 and -0.161052 values between -0.16 and -0.032190 values between -0.03 and 0.11441 values between 0.11 and 0.2437 values between 0.24 and 0.375 values between 0.37 and 0.501 values between 0.50 and 0.631 values between 0.63 and 0.771 values between 0.90 and 1.031 values between 1.16 and 1.291 values between 1.42 and 1.551 values between 2.21 and 2.341 values between 4.19 and 4.32offnegativepositiveoften

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. prend
  2. ケット
  3. 最大
  4. ボード
  5. yourself
  6. /bind
  7. ลาด
  8. 都想
  9. combin
  10. imet

Suppresses

  1. sts
  2. s
  3. atic
  4. ующ
  5. tas
  6. sti
  7. oS
  8. icated
  9. ian
  10. st