Neuron Explorer

Qwen3-0.6B

'than' in comparisons

A neuron in layer 11 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 "than" when it follows a comparison, like "faster than" or "more likely than". When it was turned up, the text drifted into repetitive or unrelated sentences, so the steering did not show a clear theme.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    either a dark black hand or a white hand. Though the cards themselves were the same, cards held in a black hand sold for about 20 percent less than cards held in a white hand. What’s more, “the race effect was

  2. #2 peak

    needs. Public charter schools are free, non-discriminatory, and do not have special entrance requirements. Ample evidence suggests that public charter schools perform better than traditional public schools on a national level, particularly for low-income students of color.

  3. #3 peak

    crashes and more than 350,000 were treated in emergency departments. Drivers aged 16 to 19 are four times more likely than older drivers to crash for every mile driven.⏎The graduated licensing programs are meant to

  4. #4 peak

    Sardinella anchovia). In some areas, the terms anchovy and sardine are used interchangeably. In addition to being slimmer and smaller than sardines,

  5. #5 peak

    between the hourly wages earned by men and women in the labor market, expressed as a percentage of men’s average wage. Women are generally considered to be paid less than men. There are two distinct numbers regarding the pay gap: unadjusted versus adjusted

  6. #6 peak

    . In China, from the 5th cent. B.C., when it was first used, calligraphy has always been considered equal, or even superior, to painting. Chinese calligraphy began with a simplified seal script, known as "ch

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 story of the House of the Seagull" is one that has always been told with a sense of mystery and intrigue. The house is a place of wonder and the Seigend's family. The story of the

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

    fairly vanilla neural networks called Alice, Bob, and Eve.⏎Ref: Learning to Protect Communications with Adversarial Neural Cryptography. arXiv - Computer Science > Cryptography and Security (21 October 2016) | ar

  2. #2 peak

    to everything we see today.⏎This composition indicates that Terzan 5 evolved its stellar populations as a dwarf galaxy, and must once have been much larger than it is now. As old stars explode in supernovae they spread newly created material huge

  3. #3 peak

    Add your thoughts below. You can sign in with Google, Twitter, Facebook, Yahoo! to get

  4. #4 peak

    transmission from infected pregnant women to their babies during pregnancy and breastfeeding while preserving the health of these children and their mothers.⏎The PROMISE (“Promoting Maternal-Infant Survival Everywhere”) study aims to enroll 7,950 HIV-inf

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. L6:1828 suppresses
  2. L6:1477 suppresses
  3. L10:2213 suppresses 'to' or 'into' showing movement between two things
  4. L6:800 excites
  5. L4:1330 suppresses
  6. L9:2267 excites 'with' after exchange nouns
  7. L9:1144 excites
  8. L10:2773 excites
  9. L9:319 excites equals signs in maths and equations
  10. L9:1449 suppresses
  11. L4:1560 suppresses
  12. L8:1319 suppresses
  13. L6:2605 suppresses
  14. L4:2069 excites
  15. L10:1329 suppresses
  16. L5:1477 excites
  17. L9:533 suppresses
  18. L6:2066 weakly excites 'of' linking a category to its parent field
  19. L10:272 weakly suppresses
  20. L3:566 weakly excites
  21. L1:895 weakly excites
  22. L10:2698 weakly excites
  23. L9:1701 weakly excites "and" joining a pair of nouns
  24. L6:1084 weakly suppresses

Feeds later layers

  1. L13:2356 excites
  2. L16:1469 excites
  3. L15:2247 excites
  4. L15:1145 excites
  5. L13:545 excites 'of' or 'to' linking two things
  6. L12:1395 excites 'as' in 'as ... as' comparisons
  7. L16:2219 excites
  8. L18:1857 excites
  9. L12:2548 excites
  10. L16:1074 excites
  11. L14:2243 excites
  12. L16:520 suppresses
  13. L14:2079 excites prepositions after comparison words
  14. L14:1421 excites
  15. L12:2721 excites
  16. L16:686 excites
  17. L16:1999 excites
  18. L16:2061 excites
  19. L16:2521 excites
  20. L15:2006 excites
  21. L12:2855 excites
  22. L16:1830 excites
  23. L17:439 excites
  24. L12:2014 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 -1.73 and -1.371 values between -1.37 and -1.025 values between -1.02 and -0.67276 values between -0.67 and -0.322998 values between -0.32 and 0.03553 values between 0.03 and 0.3926 values between 0.39 and 0.7415 values between 0.74 and 1.099 values between 1.09 and 1.444 values between 1.44 and 1.802 values between 1.80 and 2.156 values between 2.15 and 2.502 values between 2.50 and 2.851 values between 2.85 and 3.202 values between 3.20 and 3.562 values between 3.56 and 3.912 values between 4.26 and 4.611 values between 4.96 and 5.321 values between 8.84 and 9.191 values between 12.01 and 12.36offnegativepositiveoften

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. icz
  2. igos
  3. imated
  4. abei
  5. RTVF
  6. 'action
  7. 救命
  8. decess
  9. arend
  10. buch

Suppresses

  1. quotes
  2. uliar
  3. ущ
  4. fällt
  5. inclined
  6. 不完
  7. pp
  8. RING
  9. ща
  10. 곡