Neuron Explorer

Qwen3-0.6B

'and' or 'social' in academic field names

A neuron in layer 23 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

It fires on the joining word or the word 'social' inside names of academic subjects, such as 'arts and sciences', 'humanities and social sciences' and 'social sciences'. When turned up, the model kept repeating 'social sciences'.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    thoroughly up-to-date in terms of the latest developments in the field, imminently practical. It is the single most important introductory book on qualitative inquiry in the social sciences today' - Norman K. Denzin, University of Illinois.

  2. #2 peak

    by inventor, industrialist and philanthropist, Peter Cooper, offers a world-class education in art, architecture and engineering as well as an outstanding faculty of humanities and social sciences.⏎From the start, Cooper Union was a unique institution, dedicated to

  3. #3 peak

    greatest calamity.⏎The Civil War historian in Kennesaw State University’s Department of History and Philosophy recognized this void and secured the support of the College of Humanities and Social Science to begin developing the Center. After nearly a year of planning and designing

  4. #4 peak

    , 1996.⏎1919-Baylor Hospital organized in Dallas – now Baylor University Medical Center.⏎1919-College of Arts and Sciences organized.⏎1919-College of Fine Arts organized, which consisted

  5. #5 peak

    receive child care in private homes is sixteen times higher than the fatality rate for children in child care centers.⏎The study appears in the October issue of the American Sociological Review . It was funded by the Foundation for Child Development.⏎While more than

  6. #6 peak

    the rich human plurality of histories, truths, and worldviews.⏎A strong multidisciplinary foundation draws connections between and among academic and intellectual domains of the arts and sciences, as well as the philosophies and religions of the world. Humanities courses

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 I were a kid who had a super-strong voice, can I walk around around around around around... around around around around around... around around around around... around... around around... around... around... around... around... around... around

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

    ation-Resource Identification-Security-Regolith Explorer (OSIRIS-REx). Samples returned from the primitive surface of the near-Earth asteroid currently called (101955) 1999 RQ36

  2. #2 peak

    a 10-year trek through the solar system, the European Space Agency's Rosetta spacecraft has made some amazing discoveries about Comet 67P/Churyumov–Gerasimenko. From its

  3. #3 peak

    Dawn mission to Vesta and Ceres is managed by NASA’s Jet Propulsion Laboratory, a division of the California Institute of Technology in Pasadena, for NASA’s Science Mission Directorate, Washington D.C. UCLA is responsible for overall Dawn mission science.

  4. #4 peak

    but you won't be seeing this with your birdwatching binocs.⏎I don't have any images of it... yet. It's listed in the Minor Planet and Comet Ephemeris Service; put "Philplait" into the big

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. L22:1661 suppresses
  2. L20:1661 excites
  3. L22:2218 weakly suppresses
  4. L21:1186 weakly suppresses
  5. L21:1870 weakly suppresses
  6. L20:2804 weakly suppresses
  7. L2:966 weakly excites
  8. L19:2325 weakly suppresses
  9. L5:347 weakly suppresses
  10. L8:1585 weakly suppresses
  11. L1:882 weakly suppresses
  12. L10:1880 weakly suppresses
  13. L20:1819 weakly suppresses
  14. L14:1552 weakly excites
  15. L0:1404 weakly excites
  16. L16:1845 weakly suppresses
  17. L5:449 weakly excites
  18. L22:2623 weakly excites
  19. L6:1643 weakly suppresses
  20. L0:124 weakly suppresses
  21. L13:2392 weakly excites
  22. L4:1023 weakly suppresses
  23. L7:898 weakly suppresses
  24. L10:1813 weakly suppresses

Feeds later layers

  1. L26:842 excites
  2. L24:423 excites
  3. L26:2513 weakly excites
  4. L24:624 weakly excites
  5. L27:2615 weakly excites
  6. L26:2948 weakly suppresses
  7. L24:2401 weakly excites
  8. L27:2371 weakly excites
  9. L26:442 weakly excites
  10. L27:1920 weakly excites
  11. L27:2986 weakly excites
  12. L24:2270 weakly suppresses
  13. L24:2921 weakly suppresses
  14. L24:2197 weakly suppresses
  15. L26:333 weakly suppresses
  16. L24:1113 weakly excites
  17. L24:3041 weakly excites
  18. L24:1226 weakly suppresses
  19. L24:2817 weakly suppresses
  20. L25:2656 weakly excites
  21. L27:329 weakly excites
  22. L24:2505 weakly suppresses
  23. L24:1245 weakly suppresses
  24. L27:2917 weakly 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 -11.46 and -9.992 values between -9.99 and -8.523 values between -7.05 and -5.589 values between -5.58 and -4.1113 values between -4.11 and -2.6444 values between -2.64 and -1.173199 values between -1.17 and 0.31536 values between 0.31 and 1.7841 values between 1.78 and 3.2517 values between 3.25 and 4.729 values between 4.72 and 6.198 values between 6.19 and 7.664 values between 7.66 and 9.134 values between 9.13 and 10.605 values between 10.60 and 12.072 values between 12.07 and 13.541 values between 13.54 and 15.021 values between 15.02 and 16.492 values between 17.96 and 19.431 values between 20.90 and 22.371 values between 22.37 and 23.841 values between 23.84 and 25.312 values between 29.72 and 31.201 values between 34.14 and 35.611 values between 45.90 and 47.38offnegativepositiveoften

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. social
  2. natural
  3. Social
  4. social
  5. Social
  6. 社会
  7. sciences
  8. 科学
  9. 自然科学
  10. SOCIAL

Suppresses

  1. dea
  2. flate
  3. Offline
  4. itol
  5. Matter
  6. izzling
  7. Fry
  8. empor
  9. рит
  10. notifyDataSetChanged