Neuron Explorer

Qwen3-0.6B

'American' in organization names

A neuron in layer 23 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 word 'American' when it starts the name of an institution or journal, such as 'American Astronomical Society'. It works in the opposite direction on 'Center', as in 'Center for Disease Control'. Turning it up gave no 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

    over the past two years as part of the Sloan Digital Sky Survey (SDSS-II) and presented last week at the 209th meeting of the American Astronomical Society in Seattle.⏎Seven of the new galaxies are gravitationally

  2. #2 peak

    to take medicine at a certain time or getting to a doctor’s appointment. The research appears in the January issue of Neuropsychology, which is published by the American Psychological Association.⏎People with this genotype have a certain variety, or allele, of

  3. #3 peak

    Report: Dispersal of Aedes Aegypti in an Urban Area After Blood Feeding as Demonstrated by Rubidium-Marked Eggs," American Journal of Tropical Medicine Hygiene⏎52 (2):177-

  4. #4 peak

    browser, you may see unexpected results.⏎The Astronomy Center is a collection of digital resources for college-level introductory astronomy faculty and students. Astronomy Center is sponsored by the American Astronomical Society, the American Association of Physics Teachers, and the NSF-

  5. #5 peak

    50 percent more likely to develop cancer of the salivary glands, according to a new study conducted by researchers at Tel Aviv University and published in the American Journal of Epidemiology.⏎Researchers studied 500 Israelis who had tumors of

  6. #6 peak

    acid for humans. It cannot be created by the body and must be obtained from food sources.⏎The study comparing whey protein and soy protein was presented at the American College of Sports Medicine (ACSM) Annual Meeting earlier this year

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: for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for

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

    andra, two researchers from Harvard.⏎By 2020, almost one out of nine American jobs will be in health care, an April study from the Center for Health Workforce Studies projects.⏎As the country wrestles to reduce health care

  2. #2 peak

    1970 to the age of 25 in 2006.⏎Researchers T.J. Mathews and Brady E. Hamilton of the Center for Disease Control and Prevention’s (CDC) National Center for Health Statistics, Division

  3. #3 peak

    nicotine than their labels show. Or much less.⏎That's the finding of a new study by the Salt Lake County Health Department and the Center for Human Toxicology at the University of Utah, a health department news release said

  4. #4 peak

    ⏎McMurdo Ice Shelf, Antarctica© Anne D Jungblut⏎Phormidium pseudpriestleyi© Patricia Mesquita, Centre for Northern Studies (CEN)⏎Scientists sampling cyanobacteria-dominated microbial mats in

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. L21:1379 excites
  2. L20:257 weakly excites
  3. L11:2738 weakly excites
  4. L8:1327 weakly excites
  5. L20:2334 weakly suppresses
  6. L12:3058 weakly excites
  7. L21:1507 weakly excites
  8. L10:539 weakly excites
  9. L19:683 weakly suppresses
  10. L14:2743 weakly excites
  11. L20:507 weakly excites
  12. L11:2589 weakly suppresses
  13. L0:196 weakly suppresses
  14. L18:317 weakly excites
  15. L18:102 weakly suppresses
  16. L7:2194 weakly suppresses
  17. L3:1703 weakly excites
  18. L3:410 weakly excites
  19. L4:2018 weakly suppresses
  20. L8:1534 weakly excites
  21. L13:534 weakly suppresses unclear: sentence-ending periods
  22. L15:2079 weakly suppresses
  23. L5:960 weakly suppresses
  24. L18:2706 weakly excites

Feeds later layers

  1. L25:553 strongly suppresses
  2. L25:1636 strongly suppresses
  3. L26:82 strongly excites
  4. L27:463 strongly suppresses
  5. L24:1616 excites
  6. L26:890 excites
  7. L26:1527 suppresses
  8. L26:148 excites
  9. L26:884 excites
  10. L25:1843 weakly excites
  11. L24:1784 weakly excites
  12. L25:2161 weakly excites
  13. L24:1149 weakly excites
  14. L26:2540 weakly excites
  15. L25:1484 weakly suppresses
  16. L25:1164 weakly excites
  17. L27:1845 weakly excites
  18. L27:2922 weakly excites
  19. L27:1978 weakly excites
  20. L26:463 weakly excites digits inside years and numbers
  21. L27:2506 weakly excites
  22. L24:909 weakly excites
  23. L27:1187 weakly excites
  24. L26:633 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 -35.59 and -34.311 values between -16.27 and -14.981 values between -14.98 and -13.691 values between -12.41 and -11.121 values between -11.12 and -9.831 values between -9.83 and -8.542 values between -8.54 and -7.251 values between -7.25 and -5.964 values between -5.96 and -4.6812 values between -4.68 and -3.3916 values between -3.39 and -2.1038 values between -2.10 and -0.813542 values between -0.81 and 0.48257 values between 0.48 and 1.7714 values between 1.77 and 3.065 values between 3.06 and 4.344 values between 4.34 and 5.633 values between 5.63 and 6.922 values between 8.21 and 9.501 values between 10.78 and 12.071 values between 14.65 and 15.94offnegativepositiveoften

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. 玩家来说
  2. bidden
  3. calle
  4. elkaar
  5. purposes
  6. 每
  7. 每个
  8. arcpy
  9. 每位
  10. oder

Suppresses

  1. for
  2. for
  3. 对于
  4. For
  5. For
  6. →for
  7. 對於
  8. _for
  9. для
  10. _FOR