Neuron Explorer

Qwen3-0.6B

"and" inside organisation names

A neuron in layer 9 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 word "and" when it sits in the middle of a formal name, like an institute, a division or a centre ("Wage and Hour Division", "Center for Economic and Community Development"). Turning it up made the model's text drift toward unrelated, rambling topics, so the steering output doesn't 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

    even try new foods! Learn more about the role of food, health and sustainable lifestyle from the UCD Institute for Food and Health lecture series which seeks to harness the expertise of researchers to future-proof global food

  2. #2 peak

    your doctor to find out which tests are right for you.⏎Other Works Consulted⏎- Expert Panel on Integrated Guidelines for Cardiovascular Health and Risk Reduction in Children and Adolescents (2011). Expert panel on integrated guidelines for cardiovascular health

  3. #3 peak

    from one to the other. One qualification: OED refers only to a Roman vaulter but the Wordsworth Dictionary of Phrase and Fable elaborates. Not the most authoritative of sources, hence the qualification,

  4. #4 peak

    universal and regional human rights instruments such as the International Covenant on Civil and Political Rights. The African Charter, and the International Convention on Social Economical and Cultural Rights. In their preambles,the treaties ‘recall that human rights are

  5. #5 peak

    must not be interpreted as definitive evidence that a fetus does or does not have an aneuploidy. Recent recommendations from the American College of Obstetricians and Gynecologists (ACOG) are quite clear on that issue.⏎In those

  6. #6 peak

    an earthquake shook the ground, as reported in the Jakarta Globe. The Jakarta Post quoted Thomas Djamaluddin, head of the Lapan Center for Climate and Atmosphere Science Implementation as saying that the explosion was due

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 have a problem" (I have a problem) (I have a problem) (I) (I) (I) (I) (I) (I) (I) (I) (I) (I)

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

    -1842, and a letter from John Ridge to Georgia governor Wilson Lumpkin dated September 22, 1833 notifies the governor of murders committed against pro-removal Cherokees by anti-removal Ch

  2. #2 peak

    ner is an articulate and convincing salesman for his new method, and the national laboratory where he works has an excellent reputation; he is able to convince the management board of Toxic Waste Disposal Administration to use the new method. Federal funding is provided to

  3. #3 peak

    Trials network is conducting the study with funding from the National Institute of Allergy and Infectious Diseases and the Eunice Kennedy Shriver National Institute of Child Health and Human Development, both part of the National Institutes of Health. Led by protocol chair

  4. #4 peak

    1430-1480), secretary to Cardinal Bessarion and archbishop of Siponto, donated to Sassoferrato, the city of his birth, a collection of reliquaries, including the Mosaic

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. L8:982 strongly suppresses
  2. L8:419 strongly suppresses
  3. L7:223 strongly excites
  4. L3:47 strongly suppresses
  5. L7:685 excites
  6. L6:32 excites
  7. L7:489 suppresses
  8. L7:712 excites
  9. L4:152 excites
  10. L3:74 excites
  11. L8:926 excites
  12. L6:2415 suppresses
  13. L6:199 excites
  14. L7:2503 suppresses
  15. L6:181 excites periods in abbreviations and the 's in names
  16. L4:48 suppresses
  17. L7:401 suppresses
  18. L7:616 excites
  19. L8:542 excites
  20. L6:141 suppresses
  21. L8:1613 excites
  22. L7:173 excites
  23. L2:26 suppresses
  24. L8:1038 excites

Feeds later layers

  1. L12:1823 strongly suppresses
  2. L10:452 strongly suppresses
  3. L12:1112 strongly suppresses
  4. L14:493 strongly suppresses
  5. L13:1108 strongly suppresses
  6. L12:792 suppresses
  7. L10:703 excites
  8. L15:1846 suppresses
  9. L18:933 suppresses
  10. L14:347 suppresses
  11. L13:423 suppresses
  12. L17:1567 suppresses
  13. L16:928 suppresses
  14. L13:473 suppresses
  15. L12:707 suppresses
  16. L10:235 suppresses
  17. L12:940 excites
  18. L16:611 suppresses
  19. L11:1753 suppresses
  20. L15:1625 suppresses
  21. L18:617 suppresses
  22. L10:125 excites
  23. L12:3063 suppresses
  24. L10:1317 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 -4.28 and -4.091 values between -3.36 and -3.172 values between -3.17 and -2.991 values between -2.99 and -2.814 values between -2.62 and -2.441 values between -2.25 and -2.074 values between -2.07 and -1.893 values between -1.89 and -1.705 values between -1.70 and -1.529 values between -1.52 and -1.339 values between -1.33 and -1.1515 values between -1.15 and -0.9720 values between -0.97 and -0.7835 values between -0.78 and -0.6036 values between -0.60 and -0.41118 values between -0.41 and -0.23535 values between -0.23 and -0.051512 values between -0.05 and 0.14908 values between 0.14 and 0.32492 values between 0.32 and 0.51138 values between 0.51 and 0.6931 values between 0.69 and 0.8810 values between 0.88 and 1.066 values between 1.06 and 1.243 values between 1.24 and 1.432 values between 1.43 and 1.612 values between 1.61 and 1.801 values between 1.80 and 1.981 values between 2.35 and 2.531 values between 2.53 and 2.722 values between 2.90 and 3.08offnegativepositiveoften

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. 다
  2. bye
  3. ewise
  4. aul
  5. protected
  6. _coll
  7. malı
  8. ongs
  9. confirmed
  10. ustr

Suppresses

  1. =⏎⏎
  2. 性和
  3. ({⏎⏎
  4. etch
  5. "⏎⏎
  6. .";⏎⏎
  7. 期内
  8. smarty
  9. boBox
  10. .*⏎⏎