Neuron Explorer

Qwen3-0.6B

'well' after is or are

A neuron in layer 5 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

The neuron fires on the word "well" in phrases such as "is well known" or "is well within limits", where "well" follows a form of "to be". Its strongest negative cases are phrases like "quality over quantity" and "is important", which look like the opposite side of the same idiom pattern. Turning it up gave mostly off-topic text.

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

    broken peace and prosperity. The description of the king's palace with its fine furnishings, and its wonderful court containing the orchard of everbearing trees and vines is well known and is quite remarkable. The chief employment of the

  2. #2 peak

    with the larger form, but is not so fierce, and the white and pied rats kept as pets are generally varieties of this species. The habits of rats are well known. They are nocturnal bur

  3. #3 peak

    superhuman, unearthly scope, its mighty and unconquerable rhythm and unfailing light, the light of faith in eternal victory.⏎The story of Hercules is well known. His mother was a mortal, but his father was a god, the

  4. #4 peak

    body via an endoscope. The 100 millijoules per square centimetre of infrared energy that would be needed to rupture the capsules is well within safety limits. “It’s way below that used to remove<|endoftext|>Is a

  5. #5 peak

    cells can use morphogens to grow, induce a blood supply, and metastasize.⏎Although the concept of cell organization in response to these morphogen gradients is well documented, little is known about how these subtle concentration changes get established the first place

  6. #6 peak

    receives a⏎find polish, are found in this county, and large quantities of it are quarried⏎and transported to various markets. This county is admirably well⏎watered by Otter Creek, which rises near its southern boundary, and

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: a good product, and a great product. The fact that a product is always available is the same. When it comes to effectiveness, the same product is a good way to go, and the same as the value of a product. 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

    give students new experience in more realistic environment. Classes are smaller; assuring student’s best possible attention in there specialization area. It embodies the principle of “quality over quantity”.⏎These courses have smaller number of students and classes are more goal orient

  2. #2 peak

    but you can make some pretty boneheaded mistakes and still get decent results, even if your attention to detail makes darkroom perfectionists cringe. Consistency is important, but it comes naturally with experience, and in the meantime, you’ll be

  3. #3 peak

    Skin Type) have a high likelihood of developing one or more of these common precancers if they spend time in the sun and live long enough. Location makes a difference: The closer to the equator you live, the more likely you are

  4. #4 peak

    , and , they help frame the education reform debate by publicly identifying what is important for schools to teach and for students to be able to . The standards emphasize depth over breadth, building upon key concepts as students advance. The standards promote student-driven learning

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. L3:570 excites
  2. L4:881 weakly excites
  3. L3:1439 weakly excites
  4. L3:759 weakly excites
  5. L3:718 weakly suppresses
  6. L4:139 weakly excites
  7. L3:313 weakly suppresses
  8. L3:2898 weakly excites
  9. L3:527 weakly suppresses
  10. L3:2577 weakly suppresses
  11. L1:381 weakly excites
  12. L0:610 weakly excites
  13. L2:484 weakly suppresses
  14. L2:850 weakly suppresses
  15. L2:1928 weakly excites
  16. L1:496 weakly suppresses
  17. L4:2230 weakly suppresses
  18. L4:1026 weakly suppresses
  19. L1:2661 weakly suppresses
  20. L4:1379 weakly excites
  21. L4:2028 weakly suppresses
  22. L4:1157 weakly suppresses
  23. L4:2381 weakly excites
  24. L3:456 weakly suppresses

Feeds later layers

  1. L7:2155 suppresses
  2. L7:2596 suppresses
  3. L11:585 suppresses
  4. L13:2907 suppresses
  5. L7:1118 suppresses
  6. L18:2995 weakly suppresses
  7. L8:2430 weakly suppresses
  8. L22:2628 weakly excites
  9. L6:2769 weakly suppresses
  10. L7:554 weakly suppresses
  11. L10:899 weakly suppresses
  12. L21:1886 weakly suppresses
  13. L13:1270 weakly suppresses
  14. L9:806 weakly excites
  15. L13:1133 weakly suppresses
  16. L7:747 weakly suppresses
  17. L7:1191 weakly suppresses
  18. L8:2278 weakly suppresses
  19. L21:1550 weakly excites
  20. L25:1257 weakly suppresses
  21. L9:1755 weakly suppresses
  22. L17:386 weakly excites
  23. L14:917 weakly suppresses
  24. L13:2778 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 -1.91 and -1.821 values between -1.73 and -1.641 values between -1.56 and -1.471 values between -1.47 and -1.381 values between -1.29 and -1.202 values between -1.20 and -1.121 values between -1.03 and -0.942 values between -0.94 and -0.853 values between -0.85 and -0.779 values between -0.77 and -0.687 values between -0.68 and -0.5915 values between -0.59 and -0.5019 values between -0.50 and -0.4125 values between -0.41 and -0.3378 values between -0.33 and -0.24209 values between -0.24 and -0.15563 values between -0.15 and -0.061608 values between -0.06 and 0.03834 values between 0.03 and 0.11322 values between 0.11 and 0.20147 values between 0.20 and 0.2931 values between 0.29 and 0.3810 values between 0.38 and 0.467 values between 0.46 and 0.555 values between 0.55 and 0.641 values between 0.64 and 0.731 values between 0.73 and 0.821 values between 0.82 and 0.901 values between 0.90 and 0.991 values between 1.25 and 1.341 values between 1.52 and 1.60offnegativepositiveoften

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. 琇
  2. getRoot
  3. 而且还
  4. (assign
  5. Atom
  6. 砜
  7. asInstanceOf
  8. [],⏎
  9. GetInt
  10. .setSelection

Suppresses

  1. ขาด
  2. IVITY
  3. 原则
  4. ab
  5. 流水
  6. reality
  7. lay
  8. abi
  9. deeds
  10. 戏