Neuron Explorer

Qwen3-0.6B

comma after sentence-opening adverb

A neuron in layer 0 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 comma that comes right after a linking word that starts a sentence, such as Consequently, Actually, Indeed or Later. When it was turned up, the model's text did not visibly change in a way that matches this pattern.

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

    also just transition policies.⏎Consequently, in order for climate action to be successful, indigenous peoples must be seen as powerful

  2. #2 peak

    a shield.⏎Due to the widespread use of shields by all Carolingian troops, it became the most vital defensive armour for Carolingian armies.⏎Consequently, Carolingian monarchs such as Charlemagne specifically ordered different monasteries

  3. #3 peak

    differentials.⏎Actually, there has been no shortage of explanation. Going through some of the literatures,

  4. #4 peak

    of “Seeds of Deception”, in one of the first studies in the early 1990’s, rats were fed GM tomatoes. Actually, they refused to eat them, so they had to be force fed. And,

  5. #5 peak

    to industrial robots, they are programmed to operate at lower speeds and with less ”torque strength”, making them safer to use in parallel to human workers.⏎Indeed, industrial robots are often used for larger sized tasks that require bigger installations to fulfill the

  6. #6 peak

    data the supervisor of the construction was Bela Colov.⏎Later, the tower had a defensive character as well, especially in the time of the attack

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 have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately. I have been thinking about lately

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

    subsequently destroyed by Danish invaders in the 11th century.⏎The wave of religious

  2. #2 peak

    Thus, a system may consist of neural networks named Alice and Bob, and we aim

  3. #3 peak

    Thus, the increasing demand for energy and exponential surge in consumption stimulates problems such as an

  4. #4 peak

    thereby triggering a T cell immune response locally that then attacks cancer throughout the body. We

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

None: this is the first layer.

Feeds later layers

  1. L2:1283 weakly suppresses
  2. L3:1984 weakly suppresses
  3. L3:845 weakly excites
  4. L7:1776 weakly suppresses
  5. L6:427 weakly suppresses
  6. L21:1096 weakly suppresses
  7. L10:2550 weakly suppresses
  8. L6:1760 weakly suppresses data quality and data formatting
  9. L15:716 weakly excites
  10. L8:1197 weakly suppresses
  11. L18:1678 weakly excites
  12. L4:1078 weakly suppresses
  13. L5:2509 weakly excites
  14. L22:415 weakly excites
  15. L12:766 weakly suppresses
  16. L18:2548 weakly excites
  17. L24:1650 weakly suppresses
  18. L23:478 weakly suppresses
  19. L1:2069 weakly suppresses
  20. L7:1809 weakly suppresses
  21. L8:1399 weakly suppresses
  22. L7:2854 weakly suppresses
  23. L19:2514 weakly suppresses
  24. L25:1532 weakly suppresses 'New' in place names

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.

38 values between -0.63 and -0.5349 values between -0.53 and -0.4327 values between -0.43 and -0.3328 values between -0.33 and -0.2358 values between -0.23 and -0.13653 values between -0.13 and -0.022622 values between -0.02 and 0.08390 values between 0.08 and 0.1816 values between 0.18 and 0.282 values between 0.28 and 0.382 values between 0.38 and 0.481 values between 0.48 and 0.581 values between 0.58 and 0.681 values between 0.68 and 0.794 values between 0.89 and 0.991 values between 1.49 and 1.601 values between 1.60 and 1.701 values between 2.00 and 2.101 values between 2.10 and 2.204 values between 2.20 and 2.302 values between 2.30 and 2.411 values between 2.41 and 2.511 values between 2.51 and 2.611 values between 2.81 and 2.911 values between 3.22 and 3.322 values between 3.32 and 3.42offnegativepositiveoften

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. ongs
  2. 了许多
  3. StateManager
  4. ителя
  5. ителей
  6. essor
  7. ister
  8. 咛
  9. ynet
  10. .deg

Suppresses

  1. ocket
  2. 发现问题
  3. tesy
  4. 目的是
  5. **)&
  6. romise
  7. hecy
  8. 以此
  9. 셋
  10. 反应