Neuron Explorer

Qwen3-0.6B

'more' after 'a' or 'prefer a'

A neuron in layer 13 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 "more" when it comes right after "a" or "prefer a", as in "a more liberal" or "a more informal style". When it was turned up, the model's writing drifted toward sunny, descriptive sentences that did not clearly relate to the neuron's trigger.

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

    Caird, professor of moral philosophy and⏎rejected the scholastic Calvinist teaching as the true and sufficient⏎expression of the evangelical faith, instead following a more liberal⏎

  2. #2 peak

    rules of literature that are accepted, while non-formal essays deviate from this style. Some writers choose to write in a formal style while others might prefer a more informal style. There is no one style that is perfect; different writers may convey

  3. #3 peak

    variety of shapes.⏎If you want a formal looking yard, group similar plants and colors together, with spacing wide enough to allow plant distinction. If you prefer a more

  4. #4 peak

    ; only in England were the religious folk traditions of Christmas permanently altered. (It is ironic that fervent Puritan Christians began what turned into a trend to a more secular Christmas observance.) Because the common people so loved St. Nicholas, he

  5. #5 peak

    es⏎For those of you who prefer a more visual experience, here is a video review by ReviewsMS’ YouTube channel. Do

  6. #6 peak

    |Sensors and |⏎Flower-class corvettes like La Malbaie serving with the Royal Canadian Navy during the Second World War were different from earlier and more traditional sail-driven corvettes. The "corvette" designation was created by

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 set of numbers. I have a set of numbers. I have a set of numbers. I have a set of numbers. I have a set of numbers. I have a set of numbers. I have a set of numbers

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

    language, attention and impulse control. But it is still not clear what role sleep plays in the development of such brain connections.⏎In the new study, appearing online in the journal Brain Sciences, the researchers looked at differences in brain activity during sleep as

  2. #2 peak

    those who ultimately return to smoking will do so by 6 months."⏎A study published in the July 27, 2009 online version of Pediatrics involved

  3. #3 peak

    in the U.S. and Canada, according to a new study published in the journal Environmental Science and Technology Letters. In the study, the researchers tested

  4. #4 peak

    , significantly higher than the 30 to 50 percent efficiencies previously reported for other solid oxide fuel cell systems of its size, according to a study published in this month's issue of Journal of Power Sources.⏎The smaller system, developed at

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. L9:758 suppresses
  2. L11:416 suppresses
  3. L6:2443 suppresses
  4. L6:2075 suppresses
  5. L6:1269 excites
  6. L8:1572 suppresses
  7. L6:1641 suppresses
  8. L11:579 excites
  9. L12:608 excites
  10. L6:2252 excites
  11. L12:1859 excites
  12. L9:998 suppresses
  13. L11:2497 excites
  14. L11:810 excites
  15. L6:376 suppresses
  16. L6:2091 suppresses
  17. L8:537 weakly excites
  18. L11:293 weakly suppresses
  19. L11:502 weakly suppresses
  20. L11:1246 weakly excites
  21. L10:3007 weakly suppresses
  22. L5:1636 weakly excites
  23. L12:3004 weakly suppresses
  24. L9:2871 weakly excites

Feeds later layers

  1. L19:816 excites
  2. L18:1271 excites
  3. L17:57 excites
  4. L16:340 excites
  5. L18:1959 excites
  6. L19:1620 excites
  7. L17:1086 excites 'something' as an object
  8. L17:1803 excites
  9. L15:2736 excites
  10. L21:225 suppresses
  11. L14:1693 suppresses
  12. L19:237 suppresses
  13. L19:1563 excites
  14. L20:1372 excites
  15. L16:1380 excites
  16. L14:785 excites
  17. L14:1690 excites
  18. L21:1722 excites
  19. L23:718 excites
  20. L18:156 excites
  21. L16:2063 excites
  22. L18:538 excites
  23. L15:218 suppresses
  24. L14:113 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 -1.55 and -1.371 values between -1.37 and -1.202 values between -1.20 and -1.028 values between -1.02 and -0.8421 values between -0.84 and -0.6660 values between -0.66 and -0.49244 values between -0.49 and -0.31652 values between -0.31 and -0.131871 values between -0.13 and 0.05771 values between 0.05 and 0.23142 values between 0.23 and 0.4051 values between 0.40 and 0.5819 values between 0.58 and 0.7620 values between 0.76 and 0.9414 values between 0.94 and 1.115 values between 1.11 and 1.296 values between 1.29 and 1.471 values between 1.47 and 1.653 values between 1.65 and 1.823 values between 2.00 and 2.183 values between 2.36 and 2.541 values between 2.54 and 2.711 values between 2.71 and 2.891 values between 3.07 and 3.252 values between 3.60 and 3.781 values between 3.78 and 3.961 values between 3.96 and 4.131 values between 4.67 and 4.841 values between 5.02 and 5.201 values between 5.38 and 5.56offnegativepositiveoften

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. solid
  2. solid
  3. physical
  4. physiological
  5. non
  6. ocking
  7. physical
  8. non
  9. física
  10. trig

Suppresses

  1. чат
  2. atively
  3. це
  4. 提
  5. нуть
  6. ivamente
  7. Preferences
  8. ately
  9. ицы
  10. Dating