Neuron Explorer

Qwen3-0.6B

comma after an introductory phrase

A neuron in layer 12 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 reacts to the comma that comes after a short opening phrase, like 'Even years after the bullying has stopped,' right before the main point of the sentence is stated. It works in the other direction too: it dips on 'then' in 'if ... then' sentences. The steered outputs did not show a clear theme.

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

    Subsequent challenge with booster dose more effective.⏎|| Subsequent administration of antibody is less effective due to immune elimination.⏎||After antigenic stimulus, negative phase may occur due to antigen combining with any pre-existing antibody in blood.⏎

  2. #2 peak

    For those who aren’t genetically susceptible, it can cause acute illness while the biotoxins are being processed in the liver and released. However, following exposure, their detoxification system can eliminate the biotoxins via the normal mechanisms and

  3. #3 peak

    addiction is and how gambling fits in.⏎It can lead to the establishment of a powerful and difficult to break state<|endoftext|>Even before the devastating Ebola epidemic in West Africa, development agencies were highlighting that health indicators in this region were lagging the rest of

  4. #4 peak

    of suicide. Even years after the bullying has stopped, researchers have found that adults who were bullied as teens have higher levels of depression and

  5. #5 peak

    helping the baby make its way into the world.⏎Probably the Biggest Advance in Prenatal Care⏎As knowledge of the mechanics of pregnancy and birth have increased, so have the processes and systems that support them during gestation. Ultrasound was probably

  6. #6 peak

    and even the humiliation he encountered when the soldiers beat him and stripped off his clothes, and hung Him on the cross for the world to see. Even Through rejection, deprivation, persecution, and many other dealings with sinful man, sin continually tempted 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: what is it, what is it, what is it, what is it, what is it, what is it, what is it, what is it, what is it, what is it, what is it, what is it,

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

    middle class, multicultural Brooklyn neighborhood in which everyone seemed to get along. If this black family being a lodestar American family didn’t signal a shift away from racism then what would?⏎Just three years later in 1989

  2. #2 peak

    people do not believe that the choices they make will substantially improve the environment, then they are less likely to participate in activities like recycling, turning off lights when leaving

  3. #3 peak

    "allow cookies" to give you the best browsing experience possible. If you continue to use this website without changing your cookie settings or you click "Accept" below then you are consenting to this.<|endoftext|>Permanova’s knowledge centre for industrial lasers

  4. #4 peak

    contract. This can be a publicly traded stock or gold or oil or anything that is traded on the exchange. Let’s say the underlying of the contract is Microsoft then that means that our options contract represents 100 Microsoft shares.⏎There is

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. L9:1579 suppresses
  2. L6:2843 suppresses
  3. L8:410 suppresses
  4. L4:1529 excites
  5. L4:2769 suppresses
  6. L11:1399 suppresses
  7. L6:1471 excites
  8. L10:404 excites
  9. L9:298 weakly suppresses
  10. L10:1002 weakly excites
  11. L6:1829 weakly suppresses
  12. L9:306 weakly excites
  13. L3:989 weakly suppresses
  14. L6:1500 weakly excites
  15. L6:1485 weakly excites
  16. L6:1953 weakly excites
  17. L6:482 weakly suppresses
  18. L7:259 weakly suppresses
  19. L11:1569 weakly suppresses
  20. L3:771 weakly suppresses
  21. L3:520 weakly suppresses
  22. L6:3067 weakly excites
  23. L8:1195 weakly excites
  24. L6:231 weakly excites

Feeds later layers

  1. L13:958 excites
  2. L14:1538 suppresses
  3. L16:2846 suppresses
  4. L19:1507 excites
  5. L17:1448 suppresses
  6. L15:1543 suppresses
  7. L16:1886 excites
  8. L13:1375 suppresses
  9. L17:1129 excites
  10. L14:1124 suppresses
  11. L15:2554 suppresses
  12. L14:484 suppresses
  13. L15:3012 suppresses
  14. L18:1500 suppresses
  15. L15:1513 suppresses
  16. L15:1458 suppresses
  17. L19:3020 suppresses
  18. L15:2938 suppresses comma after an if-clause
  19. L13:878 excites
  20. L15:1116 suppresses
  21. L15:641 suppresses
  22. L17:845 suppresses comma after an introductory phrase
  23. L14:2717 suppresses
  24. L18:1187 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 -4.46 and -4.291 values between -3.97 and -3.801 values between -3.80 and -3.641 values between -3.48 and -3.311 values between -3.15 and -2.991 values between -2.82 and -2.663 values between -2.66 and -2.502 values between -2.50 and -2.332 values between -2.33 and -2.171 values between -2.17 and -2.011 values between -1.84 and -1.682 values between -1.68 and -1.511 values between -1.35 and -1.194 values between -1.19 and -1.024 values between -1.02 and -0.865 values between -0.86 and -0.709 values between -0.70 and -0.5317 values between -0.53 and -0.37164 values between -0.37 and -0.211166 values between -0.21 and -0.042025 values between -0.04 and 0.12377 values between 0.12 and 0.2875 values between 0.28 and 0.4523 values between 0.45 and 0.6110 values between 0.61 and 0.777 values between 0.77 and 0.942 values between 0.94 and 1.101 values between 1.27 and 1.431 values between 1.92 and 2.08offnegativepositiveoften

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. 侧结构性
  2. sort
  3. 被抓
  4. lopen
  5. 砒
  6. prise
  7. ştir
  8. unto
  9. &⏎
  10. sort

Suppresses

  1. 就会
  2. would
  3. 就
  4. 便会
  5. 就會
  6. 就意味着
  7. 辄
  8. 就越
  9. 往往会
  10. erland