Neuron Explorer

Qwen3-0.6B

verb after modal can or could

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 fires on the word right after a helper word like 'can', 'could' or 'would', especially 'be' or 'been' in passive phrases such as 'can now be used'. The negative side is weak and not clearly related. Turning it up or down gave unrelated text, so the effect is unclear.

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

    are unclean to Jews. The Maccabee family led a revolt, finally liberating Jerusalem and the Temple in 165 BC. Before God could be properly worshiped in the Temple, it had to be cleaned and dedicated. The

  2. #2 peak

    that music theory is as enjoyable as it is useful. Don’t get discouraged by the seemingly complicated written structure. With Music Theory for Dummies, understanding music has never been easier!⏎Note: CD-ROM/DVD and other supplementary materials are not included

  3. #3 peak

    many objects around us,<|endoftext|>18th Jan 12 14:01<< Go back to news⏎The kinetic force of braking can now be used in hybrid and electric vehicles as a capture store, to re-use as electric

  4. #4 peak

    needed to keep the crew alive for several months would be cut down tremendously, as sustenance could be delivered intravenously by robots. Exercise equipment would no longer be necessary, as the robots could stimulate major muscle groups periodically. Overall, the ships

  5. #5 peak

    that center around individual rights and intermediate institutions which preserved freedom and community.⏎That story falters, however, when we treat the horrors of Western history as if they were based in foreign ideas or ideas antithetical to the West.* Culture,

  6. #6 peak

    "over-plotting" and "under-plotting" because both interfere with the story telling ("General Issues and Signs of Metafiction").⏎Language can also be used to create metafiction, especially when language is used as part of an arbitrary

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: what is the value of the number 1000000000000000000000000000000000000000

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

    extremities and trunk are most susceptible.⏎It’s important to reduce your skin’s exposure to elements that damage the lipid layer and remove moisture from the skin. Some people live in homes with a low humidity. Adding a home humidifier to increase moisture

  2. #2 peak

    use Communized in a sentence. If you have trouble reading in Urdu we have also provided these meanings in Roman Urdu.⏎We have tried our level best to provide you as much detail on how to say Communized in Urdu as possible so you could

  3. #3 peak

    . The basic idea is to teach the student how to write good sentences with a variety of constructions using the proper punctuation for each particular construction. Some students will find it difficult if they haven't some familiarity with grammar it moves rather quickly. The first

  4. #4 peak

    see them – paper planners, calendars on the phone, everything laid out. While some people and teens are natural planners, most are not. And they may find it challenging to figure out how to “become” a planner.⏎As the saying

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. L6:960 excites
  2. L10:616 excites
  3. L8:403 weakly excites
  4. L3:2260 weakly suppresses
  5. L10:759 weakly excites
  6. L6:1269 weakly excites
  7. L10:87 weakly suppresses
  8. L10:1336 weakly excites
  9. L4:116 weakly suppresses
  10. L10:372 weakly suppresses
  11. L8:879 weakly excites
  12. L7:1701 weakly excites
  13. L6:2492 weakly suppresses
  14. L0:1918 weakly excites
  15. L1:301 weakly suppresses
  16. L7:767 weakly excites
  17. L9:1563 weakly excites
  18. L3:94 weakly suppresses
  19. L5:453 weakly excites
  20. L6:82 weakly suppresses
  21. L8:2130 weakly suppresses
  22. L2:2634 weakly excites
  23. L4:404 weakly suppresses
  24. L6:68 weakly excites

Feeds later layers

  1. L15:63 excites
  2. L13:2687 suppresses
  3. L15:1353 weakly excites
  4. L14:1857 weakly suppresses
  5. L21:1035 weakly suppresses
  6. L17:3050 weakly suppresses
  7. L15:456 weakly suppresses 'and' between two verbs
  8. L14:1519 weakly suppresses
  9. L16:529 weakly suppresses
  10. L16:2600 weakly excites
  11. L25:1741 weakly excites
  12. L22:1032 weakly excites
  13. L13:1800 weakly suppresses
  14. L15:2098 weakly excites
  15. L22:1743 weakly suppresses
  16. L12:1222 weakly suppresses
  17. L15:2682 weakly suppresses
  18. L13:895 weakly suppresses
  19. L12:2239 weakly suppresses
  20. L12:1806 weakly excites
  21. L25:969 weakly excites
  22. L23:473 weakly excites
  23. L12:2850 weakly suppresses
  24. L16:2826 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 -2.08 and -1.941 values between -1.81 and -1.675 values between -1.67 and -1.543 values between -1.54 and -1.414 values between -1.41 and -1.276 values between -1.27 and -1.1414 values between -1.14 and -1.0010 values between -1.00 and -0.8718 values between -0.87 and -0.7327 values between -0.73 and -0.6045 values between -0.60 and -0.47102 values between -0.47 and -0.33207 values between -0.33 and -0.20635 values between -0.20 and -0.061937 values between -0.06 and 0.07442 values between 0.07 and 0.21137 values between 0.21 and 0.3485 values between 0.34 and 0.4853 values between 0.48 and 0.6138 values between 0.61 and 0.7432 values between 0.74 and 0.8821 values between 0.88 and 1.0117 values between 1.01 and 1.1512 values between 1.15 and 1.2812 values between 1.28 and 1.429 values between 1.42 and 1.557 values between 1.55 and 1.685 values between 1.68 and 1.824 values between 1.82 and 1.954 values between 2.09 and 2.221 values between 2.22 and 2.365 values between 2.36 and 2.493 values between 2.49 and 2.622 values between 2.62 and 2.761 values between 2.89 and 3.031 values between 3.03 and 3.162 values between 3.16 and 3.30offnegativepositiveoften

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. })"⏎
  2. ambia
  3. _THAT
  4. iom
  5. ChangedEventArgs
  6. valido
  7. otty
  8. ))"⏎
  9. -INF
  10. ******************************************************************************/⏎

Suppresses

  1. шки
  2. (non
  3. edic
  4. チャー
  5. ded
  6. .setSelection
  7. cho
  8. handles
  9. шек
  10. 萑