Neuron Explorer

Qwen3-0.6B

page-number hyphens in citations

A neuron in layer 14 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 fires on the dash between two page numbers in a reference list, such as 'pages 1619-1634'. It hardly fires on anything else that matters. Turning it up did not produce any clear citation-like 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

    Complications Not Increased with Fewer Antibiotics⏎Am Fam Physician. 2006 May 1;73(9):1619-1634.⏎Clinical Question: Does more cautious prescribing of antibiotics for sore

  2. #2 peak

    an-RM⏎Am Ind Hyg Assoc J 1980 Oct; 41(10):758-765⏎The effectiveness of open versus closed-face cassettes in aeros

  3. #3 peak

    , CT scan, rule, minor head injury⏎Ann Afr Surg. 2018;15(2):57-61 DOI:http://dx.doi.org/10.431

  4. #4 peak

    , and rheumatic fever in children: time trend analysis. BMJ. August 6, 2005;331:328–9.⏎Used with permission from

  5. #5 peak

    W.E. (1995). Using groundwater for hydro electric power generation. Journal of Hydrology (NZ), 34(1), 1-14.⏎Permanent Research Commons link: http://hdl.handle.net/10

  6. #6 peak

    posed to gestational diabetes mellitus in utero. Diabetes Metab Syndr. 2019;13(5):2947-2952.<|endoftext|>- Published on Tuesday, 18 September

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: that the in a market economy, the only thing that the business is the only thing that the business and the business is the only thing that the business is the only thing that the business is the only thing that the business is the only thing

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

    of the work bells and the noise was too much for some. One worker left the following anonymous note when she left:⏎“I am going home where I shall not be obliged to rise so early

  2. #2 peak

    “proven safe”. They did the same with genetic engineering.⏎I’m pleased to see the Commissioner (who has not been shy to seriously attack the Government when she thinks they have it wrong), come up with sensible conclusions, and her further work

  3. #3 peak

    New York Times, Bureau of Labor Statistics⏎For Rampell, the figure makes a slam-dunk case for the financial benefits of graduating from college. Here is her take on what the graph shows:⏎The median weekly earnings for college graduates are $

  4. #4 peak

    Population Decline<|endoftext|>First posted on Myth & Moor in 2015.⏎"In Britain there were once hundreds of carnivals," writes Jay Griffiths in her excellent book Pip, Pip, a cultural

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. L10:587 strongly suppresses
  2. L6:1156 strongly suppresses
  3. L7:487 strongly suppresses
  4. L4:2396 suppresses
  5. L10:489 suppresses
  6. L7:477 suppresses
  7. L7:2804 excites
  8. L4:2471 excites
  9. L9:521 suppresses
  10. L2:2339 suppresses
  11. L4:1314 excites
  12. L8:1239 suppresses
  13. L9:472 excites
  14. L2:1754 excites
  15. L4:911 excites
  16. L8:759 excites
  17. L2:1644 suppresses
  18. L1:2869 suppresses
  19. L0:1296 suppresses
  20. L9:2071 excites
  21. L10:898 excites
  22. L3:2226 weakly excites
  23. L1:2878 weakly suppresses
  24. L10:2116 weakly excites

Feeds later layers

  1. L21:1614 strongly suppresses
  2. L17:648 strongly suppresses
  3. L22:520 strongly suppresses
  4. L18:838 strongly suppresses
  5. L21:1021 excites
  6. L20:2754 suppresses
  7. L18:1594 excites
  8. L18:2226 suppresses
  9. L25:2038 suppresses
  10. L17:287 excites
  11. L24:1888 suppresses
  12. L27:737 suppresses
  13. L24:614 suppresses
  14. L16:1347 suppresses
  15. L22:687 suppresses
  16. L15:2172 excites
  17. L21:2081 excites
  18. L26:2449 suppresses
  19. L21:2560 excites
  20. L25:1075 weakly suppresses
  21. L23:421 weakly suppresses
  22. L23:2856 weakly suppresses
  23. L19:1375 weakly excites
  24. L23:1052 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 -7.98 and -7.762 values between -4.68 and -4.461 values between -3.80 and -3.582 values between -3.58 and -3.361 values between -2.92 and -2.711 values between -2.71 and -2.491 values between -2.49 and -2.271 values between -2.05 and -1.832 values between -1.61 and -1.392 values between -1.39 and -1.177 values between -1.17 and -0.956 values between -0.95 and -0.739 values between -0.73 and -0.5155 values between -0.51 and -0.29560 values between -0.29 and -0.072490 values between -0.07 and 0.15632 values between 0.15 and 0.37115 values between 0.37 and 0.5920 values between 0.59 and 0.81offnegativepositiveoften

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. itus
  2. icho
  3. ATO
  4. Ạ
  5. \Builder
  6. medios
  7. bable
  8. TODO
  9. ->{$
  10. incurred

Suppresses

  1. maids
  2. 她们
  3. 嫁给
  4. adora
  5. arians
  6. jection
  7. jections
  8. herself
  9. 妮
  10. ари