Neuron Explorer

Qwen3-0.6B

"but" after a negated clause

A neuron in layer 1 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 "but" when it comes right after a "not X," phrase, as in "not X, but Y". Turning it up gave text that did not follow this pattern, so the steering adds little to the explanation.

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

    running. Slow to<|endoftext|>South-eastern Australia is one of the most bushfire-prone areas in the world.⏎The risk of bushfires cannot be eliminated, but their impact can be reduced by effective planning and bushfire management.⏎The National Parks

  2. #2 peak

    ask for the Negro is not benevolence, not pity, not sympathy, but simply justice.”⏎The 13th Amendment, ratified later in 18

  3. #3 peak

    as well as for psychological problems such as anxiety and panic. People often misunderstand the goal of therapy and especially mindfulness. Relaxation and happiness are not the aim, but rather a "freedom from the tendency to get drawn into automatic reactions to thoughts

  4. #4 peak

    that many jumping spiders, crab spiders, wolf spiders and other garden spiders are not indiscriminate diners, but in fact develop rather narrow tastes in prey. For example, a crab spider may

  5. #5 peak

    Rabbi Emil Fackenheim in his book, "God's Presence in History." "Nevertheless, not messengers, not angels, not intermediaries, but God himself acts in human history."⏎God hears the cries of the people Israel and

  6. #6 peak

    use of zero tolerance in schools since its inception in the 1980s. It found that not only does zero tolerance not achieve its stated goals, but that its most common punishments, suspension and expulsion, lead to increased drop

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 need to be sure I will be in a place that I need to be sure I will be in a place that I need to be sure I will be in a place that I need to be sure I will be in a place that

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

    required for digestion of food, exercise, cleaning the house, everything. Definition of terms that create your total calorie burn:⏎- RMR=resting metabolic rate-the amount of calories you need to exist without any activity-ie:bedrest⏎-

  2. #2 peak

    Roosevelt dedicated the United States to the Policy of the Good Neighbor-"the neighbor who resolutely respects himself and, because he does so, respects the rights of others-the neighbor who respects his obligations and respects the sanctity of his agreements in and with

  3. #3 peak

    be 10 0 10 why am I getting 10 0 20.I am confused can some one help me in this regard.thanks in advance. kamesh. here is the code from Marcus Mock exam

  4. #4 peak

    respect to the horizontal. (seaplane) trim-The angle with the horizontal surface of the water assumed by the float or hull under given conditions. trimming moment-The moment about a reference point, usually the center of gravity, exerted by the

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. L0:972 weakly excites
  2. L0:386 weakly suppresses
  3. L0:148 weakly suppresses
  4. L0:1513 weakly suppresses
  5. L0:84 weakly suppresses
  6. L0:1030 weakly excites
  7. L0:1721 weakly excites
  8. L0:463 weakly excites
  9. L0:606 weakly suppresses
  10. L0:97 weakly excites
  11. L0:1855 weakly suppresses
  12. L0:151 weakly suppresses
  13. L0:893 weakly excites
  14. L0:2389 weakly excites
  15. L0:1038 weakly suppresses
  16. L0:2808 weakly excites
  17. L0:1501 weakly excites
  18. L0:1915 weakly excites
  19. L0:1111 weakly excites
  20. L0:1790 weakly suppresses
  21. L0:490 weakly suppresses
  22. L0:441 weakly suppresses
  23. L0:2989 weakly excites
  24. L0:1713 weakly excites

Feeds later layers

  1. L2:829 suppresses
  2. L2:163 excites
  3. L2:263 excites
  4. L2:179 suppresses comma after a clause subject
  5. L25:1655 weakly excites
  6. L13:2141 weakly excites end-of-text token after finished articles
  7. L24:667 weakly excites
  8. L9:160 weakly suppresses
  9. L15:1959 weakly suppresses
  10. L10:492 weakly excites
  11. L27:1961 weakly excites
  12. L12:1206 weakly excites
  13. L20:850 weakly suppresses
  14. L10:144 weakly suppresses
  15. L24:1653 weakly excites
  16. L6:165 weakly excites
  17. L27:1664 weakly excites
  18. L27:2595 weakly excites
  19. L14:2459 weakly excites
  20. L27:785 weakly excites
  21. L10:650 weakly excites
  22. L4:2519 weakly excites
  23. L12:1485 weakly excites
  24. L8:613 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 -1.36 and -1.291 values between -0.82 and -0.751 values between -0.75 and -0.681 values between -0.68 and -0.611 values between -0.61 and -0.543 values between -0.54 and -0.481 values between -0.48 and -0.415 values between -0.41 and -0.3414 values between -0.34 and -0.2756 values between -0.27 and -0.20135 values between -0.20 and -0.14411 values between -0.14 and -0.071064 values between -0.07 and 0.001055 values between 0.00 and 0.07568 values between 0.07 and 0.14312 values between 0.14 and 0.21159 values between 0.21 and 0.2765 values between 0.27 and 0.3421 values between 0.34 and 0.4114 values between 0.41 and 0.487 values between 0.48 and 0.554 values between 0.55 and 0.622 values between 0.62 and 0.681 values between 0.68 and 0.754 values between 0.75 and 0.821 values between 0.89 and 0.961 values between 1.30 and 1.36offnegativepositiveoften

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. 放射
  2. 綦
  3. -secret
  4. SED
  5. Ề
  6. sez
  7. singly
  8. ировки
  9. 三位一体
  10. 𝑉

Suppresses

  1. ect
  2. 比分
  3. eur
  4. egot
  5. iger
  6. QUIRE
  7. .cgColor
  8. ftype
  9. autical
  10. �认