Neuron Explorer

Qwen3-0.6B

punctuation that introduces an explanation

A neuron in layer 18 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron lights up on a colon, dash or comma that comes right after a statement and sets up an explanation, definition or list of what follows. When it was turned up, the model's writing drifted into repetitive, rambling loops rather than anything on one topic.

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

    is used in relation to two similar concepts: identity from a recognition stand point, and identity from an authentication and authorization view. The first, is our online presence: accounts, digital footprints, what we do online, who we connect with and

  2. #2 peak

    ,” said lead author Dr. Sarah Jackson from UCL Institute of Epidemiology & Health Care.⏎“However further research is required to clarify the direction of causation — it could be the case that depression causes people to lose their interest in eating chocolate

  3. #3 peak

    can overwhelm a child and lead him down a challenging path. And, peer pressure is a real thing – it is not only observable, but changes the way the brain behaves.⏎As a

  4. #4 peak

    explain, the legal concept of bias goes much further than the factual question of whether a person is biased or not. In the law, bias is all about perception – it is a question of whether the public think that a person is biased, and

  5. #5 peak

    and sound of their parents, and may get sad when they don't see them.⏎- Step 6: Look for cause and effect recognition at 3 months, as the child picks up and drops things, using their hands to shake a r

  6. #6 peak

    all the behind-the-scenes processes that impact an organization’s bottom line.⏎In work management, professionals are concerned with the how questions: How can we increase the success of our products/projects? How can we improve our

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: the following two. 1. 1.5-1.7 and 8.3-9.3 and 10.1-11.1**1.4** 1.8 1.0

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

    eroscopy. Using a viewing instrument inserted through the vagina, your healthcare provider can see the cervix and the inside of the uterus.⏎- Laparoscopy. This minor procedure uses a laparoscope, a viewing instrument with a light,

  2. #2 peak

    of the full functionality of CTS.⏎- Common Language Run-time (CLR):⏎The core of the .NET platform is the Common Language Run-time, aka CLR. It do all the heart tasks of locating, loading and managing .NET types.

  3. #3 peak

    for Alzheimer’s.⏎- Managing stress– Stress can affect both your physical and mental health. Physically it can make conditions such as high blood pressure worse. Mentally it can cause anxiety or depression, and reduce your ability to focus attention and remember

  4. #4 peak

    the response given by arteries to the flow of blood.⏎Understanding Systolic and Diastolic Blood Pressure⏎Blood pressure is one of the vital signs of life. It is normally expressed as two numbers, the first or the upper number indicating 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. L16:3056 excites
  2. L17:2249 suppresses
  3. L14:2610 suppresses
  4. L15:2868 excites
  5. L15:1308 suppresses
  6. L13:2492 suppresses
  7. L16:398 excites
  8. L16:1394 excites
  9. L14:3066 suppresses
  10. L13:1694 excites
  11. L12:2539 suppresses
  12. L16:1652 excites
  13. L16:2933 weakly suppresses
  14. L15:2937 weakly excites
  15. L15:2670 weakly excites
  16. L14:783 weakly suppresses
  17. L12:1934 weakly suppresses
  18. L16:2485 weakly suppresses
  19. L17:750 weakly excites
  20. L12:569 weakly excites
  21. L12:2279 weakly suppresses
  22. L15:1833 weakly suppresses unclear: mixed mid-sentence words
  23. L13:2607 weakly excites
  24. L17:326 weakly suppresses

Feeds later layers

  1. L19:1469 excites
  2. L19:376 suppresses
  3. L23:594 excites
  4. L20:1007 excites
  5. L20:2161 excites
  6. L21:2485 weakly excites
  7. L19:2292 weakly excites
  8. L22:974 weakly excites
  9. L22:1592 weakly excites
  10. L19:2359 weakly excites
  11. L20:2229 weakly excites
  12. L19:2737 weakly excites
  13. L23:2329 weakly excites
  14. L19:1371 weakly excites
  15. L26:400 weakly excites
  16. L22:1407 weakly suppresses
  17. L19:2762 weakly excites
  18. L23:2774 weakly excites
  19. L20:1243 weakly excites
  20. L19:2187 weakly excites
  21. L20:837 weakly excites
  22. L21:761 weakly excites
  23. L21:655 weakly excites
  24. L20:2836 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 -3.24 and -3.052 values between -3.05 and -2.861 values between -2.86 and -2.675 values between -2.48 and -2.292 values between -2.29 and -2.106 values between -2.10 and -1.918 values between -1.91 and -1.7213 values between -1.72 and -1.5315 values between -1.53 and -1.3423 values between -1.34 and -1.1532 values between -1.15 and -0.9653 values between -0.96 and -0.7789 values between -0.77 and -0.58138 values between -0.58 and -0.39276 values between -0.39 and -0.20888 values between -0.20 and -0.011485 values between -0.01 and 0.18603 values between 0.18 and 0.37172 values between 0.37 and 0.5643 values between 0.56 and 0.7523 values between 0.75 and 0.9410 values between 0.94 and 1.133 values between 1.13 and 1.324 values between 1.32 and 1.512 values between 1.51 and 1.703 values between 1.70 and 1.893 values between 1.89 and 2.082 values between 2.46 and 2.651 values between 2.84 and 3.032 values between 4.17 and 4.36offnegativepositiveoften

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. 何时
  2. how
  3. how
  4. 谁
  5. 什么时候
  6. 眼看
  7. 如何
  8. 怎样
  9. 誰
  10. 是谁

Suppresses

  1. ipation
  2. enerative
  3. agination
  4. ROLE
  5. Agility
  6. versible
  7. ]^
  8. segregation
  9. 弃
  10. UNCTION