Neuron Explorer

Qwen3-0.6B

"even" before a comparative

A neuron in layer 8 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 "even" when it comes before "more", "worse" or "greater", as in "even more roadblocks". The steering results were unrelated to this, so the link is only seen in the examples.

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

    in science, technology, engineering, and math (STEM) positions is an ongoing concern. And the challenges brought on by the unprecedented COVID-19 pandemic created even more roadblocks for many of us.⏎According to a survey from Environment and Climate

  2. #2 peak

    , they have received comparatively little attention from taxonomists. Had it not been for their importance as parasitoids of insect pests, the group would have received even less attention. Yet evidence from the British fauna and detailed work in tropical America suggests

  3. #3 peak

    tremendous strain on electrical systems, as seen during Texas’s deadly winter storm. Summer blackouts may become even more common, and deadlier, as temperatures soar. The energy required to save

  4. #4 peak

    alusia exported olive oil to Italy and other Mediterranean countries during this era. When Andalusia later fell under Islamic rule, olive oil cultivation expanded and became an even more vital part of the regional economy.⏎Today the land is known for its iconic

  5. #5 peak

    of homes in predominantly white neighborhoods has appreciated considerably in the last 50 or so years. This development has a dual effect: It makes homes in those neighborhoods even more unattainable for African Americans and it entrenches the intergener

  6. #6 peak

    arcus, Forsyth, Stone, Dubbert, McKenzie, Dunn & Blair 2000) This lack of physical activity among the elderly is of even greater concern when considering the changing demographics.⏎The intention to perform a behavior is the

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: the first step in the game of math, addition, and the sum of the numbers formed by two numbers to the sum of two numbers to the 200000000000000000

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

    possibly represents a significant part of the future of organic chemistry with its potential role in developing, for example, man-made molecular switches or molecular muscles. The work on rotax

  2. #2 peak

    : 2011-10-14⏎Credited as : manufacturer, blue jeans, Levi Strauss & Co.⏎Levi Strauss made and sold blue jeans in San Francisco, California, during and after the gold rush.

  3. #3 peak

    September 27, 1975. The Privacy Act is essentially an omnibus "code of fair information practices" that attempts to regulate the collection, maintenance, use and dissemination of personal information by federal executive branch agencies. These amendments

  4. #4 peak

    is up to, and she slams the window sash on his beard. Trapped, he must

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. L6:2252 weakly excites
  2. L2:1655 weakly excites
  3. L6:2075 weakly suppresses
  4. L5:2319 weakly excites
  5. L0:1239 weakly excites
  6. L3:2483 weakly suppresses
  7. L2:500 weakly suppresses
  8. L7:1449 weakly excites
  9. L6:1161 weakly excites
  10. L7:400 weakly excites
  11. L3:57 weakly excites
  12. L5:2113 weakly suppresses
  13. L1:1374 weakly excites
  14. L7:1978 weakly excites
  15. L6:1066 weakly excites
  16. L1:352 weakly excites
  17. L3:885 weakly suppresses
  18. L0:332 weakly suppresses
  19. L1:1576 weakly suppresses
  20. L7:2013 weakly suppresses
  21. L4:2571 weakly suppresses
  22. L1:2552 weakly suppresses
  23. L6:1892 weakly excites
  24. L5:1134 weakly excites

Feeds later layers

  1. L9:758 excites
  2. L14:1337 excites
  3. L17:2660 excites
  4. L10:991 weakly excites
  5. L9:434 weakly excites comma after an adjective in a list
  6. L19:2194 weakly excites first half of hyphenated compound adjectives
  7. L24:2855 weakly excites
  8. L15:461 weakly excites
  9. L23:1726 weakly excites
  10. L20:1976 weakly excites
  11. L14:1332 weakly suppresses
  12. L18:1456 weakly excites
  13. L27:153 weakly suppresses the spot just before a historical number
  14. L9:707 weakly excites
  15. L23:366 weakly excites
  16. L15:1959 weakly suppresses
  17. L16:1406 weakly excites
  18. L17:968 weakly excites
  19. L16:1819 weakly excites
  20. L15:778 weakly excites
  21. L20:1979 weakly excites
  22. L19:1956 weakly suppresses
  23. L18:551 weakly suppresses
  24. L23:283 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 -1.31 and -1.163 values between -1.16 and -1.013 values between -1.01 and -0.861 values between -0.86 and -0.712 values between -0.71 and -0.5610 values between -0.56 and -0.41131 values between -0.41 and -0.26722 values between -0.26 and -0.111823 values between -0.11 and 0.04938 values between 0.04 and 0.19192 values between 0.19 and 0.3349 values between 0.33 and 0.4817 values between 0.48 and 0.632 values between 0.63 and 0.782 values between 0.78 and 0.932 values between 0.93 and 1.081 values between 1.08 and 1.233 values between 1.23 and 1.382 values between 1.38 and 1.531 values between 1.82 and 1.971 values between 3.46 and 3.611 values between 3.61 and 3.761 values between 4.50 and 4.65offnegativepositiveoften

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. ಠ
  2. aveled
  3. -headed
  4. JR
  5. obsolete
  6. sé
  7. 善意
  8. achable
  9. Writable
  10. 过剩

Suppresses

  1. imum
  2. 安阳
  3. ippi
  4. ницы
  5. isor
  6. erg
  7. programma
  8. ients
  9. яти
  10. enheim