Neuron Explorer

Qwen3-0.6B

gender and race in lists of identities

A neuron in layer 11 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

This neuron fires on the word that names a group of people, especially in lists like race, gender, religion or nationality. When it is turned up, the model's writing drifts away from the topic and does not stay on 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

    minorities has a long and storied history. Organizations that strive for equality for all races, genders, and religions can trace their origins back to a small gathering in upstate New

  2. #2 peak

    inherently entitled. It is essentially an acknowledgment that we should remember and understand the value of each human being, regardless of their nationality, color, race, sex, gender, country of origin, language, or otherwise.⏎As the world becomes more and

  3. #3 peak

    .⏎- Excessive screen time is linked to lower grades in school, reading fewer books, exercising less, and being overweight.⏎- Violence, sexuality, race and gender stereotypes, drug and alcohol abuse are common themes of television, movie and video games

  4. #4 peak

    experiences, rationality, other people, race, gender, socio-economic status, ethnicity, and religion?⏎- Are they conscious of themselves

  5. #5 peak

    across race/ethnicity subgroups between and within states when life expectancy was examined by race/ethnicity

  6. #6 peak

    on their race, ethnic background, citizenship status or national origin. This includes people of all races and ethnicities.⏎California law prohibits discrimination based on an individual’s gender, including transgender and non-binary individuals. Gender identity and gender expression are also protected

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 been given the opportunity to work on the research to be able to address the problems that are in the real world. For instance, I have been given the opportunity to study the different parts of the brain, such as the visual system

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

    ,” the DG stated.⏎“The Delta strain is spreading like wildfire over the world, causing a new surge in cases and deaths.⏎“However, not everyone is suffering equally; we are

  2. #2 peak

    without a usual source of medical care were less likely to be screened for tobacco use and receive advice to quit.⏎Members of several racial and ethnic minority populations bear a disproportionate share of the adverse health consequences of tobacco use. There is strong evidence that interventions

  3. #3 peak

    . Kathleen Cravero, deputy executive director of UNAIDS, and Ambassador George Moose, former assistant secretary of state for African affairs, shed light on the disproportionate impact HIV/AIDS has had on women in developed and developing countries. The second briefing

  4. #4 peak

    little to salvage for homeowners caught in harm's way.⏎Climate change will disproportionately affect California’s most disadvantaged and isolated communities, according to a recent report from the

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. L10:1738 suppresses
  2. L7:1124 suppresses
  3. L10:2161 excites
  4. L7:2733 excites
  5. L10:2581 suppresses
  6. L6:1964 suppresses
  7. L9:896 weakly excites
  8. L8:3053 weakly excites
  9. L4:1038 weakly excites
  10. L10:1553 weakly excites
  11. L5:2903 weakly suppresses
  12. L1:2497 weakly excites
  13. L4:935 weakly suppresses
  14. L4:1181 weakly suppresses
  15. L6:596 weakly excites
  16. L6:831 weakly suppresses
  17. L10:345 weakly suppresses
  18. L6:13 weakly excites
  19. L9:2860 weakly suppresses
  20. L1:824 weakly excites
  21. L8:1494 weakly excites preposition right after a passive verb
  22. L5:1548 weakly excites
  23. L10:1938 weakly excites
  24. L9:2509 weakly excites

Feeds later layers

  1. L14:1963 suppresses
  2. L14:2967 suppresses
  3. L23:784 suppresses
  4. L13:20 suppresses
  5. L14:1706 suppresses
  6. L13:951 suppresses
  7. L17:1600 suppresses
  8. L12:1162 suppresses
  9. L15:1952 suppresses
  10. L20:2383 suppresses
  11. L14:779 suppresses 'than' in comparisons
  12. L17:680 suppresses
  13. L19:2295 suppresses
  14. L19:2922 suppresses
  15. L21:2915 suppresses
  16. L12:2502 weakly suppresses
  17. L14:3012 weakly suppresses
  18. L12:1744 weakly suppresses
  19. L17:2700 weakly suppresses
  20. L14:365 weakly suppresses
  21. L14:253 weakly excites
  22. L21:1607 weakly suppresses
  23. L17:2904 weakly suppresses
  24. L20:1765 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 -4.13 and -3.961 values between -3.62 and -3.451 values between -2.94 and -2.771 values between -2.60 and -2.432 values between -2.27 and -2.101 values between -2.10 and -1.931 values between -1.93 and -1.762 values between -1.76 and -1.592 values between -1.59 and -1.424 values between -1.42 and -1.251 values between -1.25 and -1.086 values between -1.08 and -0.915 values between -0.91 and -0.7410 values between -0.74 and -0.5725 values between -0.57 and -0.40124 values between -0.40 and -0.23679 values between -0.23 and -0.061836 values between -0.06 and 0.11925 values between 0.11 and 0.28241 values between 0.28 and 0.4531 values between 0.45 and 0.625 values between 0.62 and 0.792 values between 0.79 and 0.961 values between 1.63 and 1.801 values between 2.48 and 2.65offnegativepositiveoften

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. 小时前
  2. eme
  3. issa
  4. bildung
  5. comed
  6. Compose
  7. .Combine
  8. referee
  9. ague
  10. ompiler

Suppresses

  1. disproportionately
  2. uneven
  3. unequal
  4. respons
  5. asym
  6. privile
  7. /gtest
  8. icism
  9. 不公平
  10. disproportionate