Neuron Explorer

Qwen3-0.6B

unclear: mid-clause words in expository prose

A neuron in layer 12 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

The neuron fires on words in the middle of explanatory sentences, like 'first' after 'hits the ground', or 'quintile' and 'average'. No single topic stands out. Turning it up made the writing drift into lists and repeated phrases, which does not reveal a clear meaning.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

    the ‘villainous molecule’ in the greenhouse effect and noted “adding carbon dioxide to the atmosphere inevitably causes a temperature rise. And the more you add, the greater the temperature rise.” (p.52) As we have noted

  2. #2 peak

    believed to act differently on different objects.⏎- Drop a metal bar and a feather at the same time… which one hits the ground first?⏎- Obviously, common sense will tell you that the bar will hit first,

  3. #3 peak

    .⏎The researchers found that higher DASH scores were associated with higher scores for cognitive functioning at the beginning of the study and over time. Those in the highest quintile scored 1.42 points higher at baseline and 1.81

  4. #4 peak

    fires to causing the least number of bushfires, and nominates which of these causes are most preventable.⏎

  5. #5 peak

    that can be performed in space. The winner/winning team’s experiment will be conducted aboard the International Space Station (ISS), making it the

  6. #6 peak

    present in the abdomen, chest, and/or under the skin⏎- Inflammation of the pancreas (pancreatitis)⏎- Shock and low levels of oxygen in the blood and tissues (hypoxia)⏎- Thrombocyt

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 idea that, as long as I am the leader, I am the leader of the nation, the country, and the common. I am the leader, I am the leader, I am the leader of the People's Republic of China

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

    the area for months. And as is common in the Desert Southwest during summer, a monsoonal flow drew low pressure into the area, leading to the development of thunderstorms early in the morning of

  2. #2 peak

    the creation of chaebol (large conglomerates), administered credit allocation with easy monetary policies from banks to priority sectors and export-oriented outward-looking policies to concentrate economic power highly

  3. #3 peak

    Plato and Plotinius - or through story - in the case of Homer.⏎Plotinus, having been a proponent of Platonism, naturally concurs with Plato, postulating that each of us is comprised of an everyday, embodied experience

  4. #4 peak

    coincidences. the roots of this philosophy go back as far as antiquity and ancient Greece.⏎All atheist philosophies that deny creation, directly or indirectly embrace and defend the idea of evolution. the same condition today applies to all the ideologies and systems

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. L10:1492 excites
  2. L11:726 suppresses
  3. L11:1959 excites
  4. L11:338 excites
  5. L11:1725 suppresses
  6. L10:1252 suppresses
  7. L11:1409 suppresses
  8. L11:2045 excites
  9. L11:1714 suppresses
  10. L11:2250 suppresses
  11. L11:743 excites
  12. L11:797 excites
  13. L11:1314 suppresses
  14. L6:2801 excites
  15. L9:2498 suppresses
  16. L11:1963 suppresses
  17. L10:2704 suppresses
  18. L9:2376 suppresses
  19. L11:2571 weakly excites
  20. L10:1510 weakly suppresses
  21. L11:1616 weakly excites
  22. L10:1257 weakly suppresses
  23. L11:1097 weakly excites
  24. L10:495 weakly excites

Feeds later layers

  1. L18:387 suppresses
  2. L15:2543 excites
  3. L15:1781 suppresses
  4. L17:2038 excites
  5. L13:2183 excites
  6. L17:1932 excites
  7. L14:1738 excites
  8. L13:1882 excites
  9. L14:2767 excites
  10. L13:459 suppresses
  11. L21:892 suppresses
  12. L14:2967 suppresses
  13. L14:1083 excites
  14. L18:3017 excites
  15. L13:1011 excites
  16. L13:1500 suppresses
  17. L16:2402 excites
  18. L13:1573 suppresses
  19. L17:1170 suppresses
  20. L14:899 suppresses
  21. L13:2558 suppresses unclear: quantities and measured changes
  22. L13:2759 excites
  23. L17:1619 suppresses
  24. L13:1366 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 -2.96 and -2.842 values between -1.72 and -1.602 values between -1.60 and -1.472 values between -1.47 and -1.355 values between -1.35 and -1.224 values between -1.22 and -1.107 values between -1.10 and -0.986 values between -0.98 and -0.8512 values between -0.85 and -0.7332 values between -0.73 and -0.6025 values between -0.60 and -0.4861 values between -0.48 and -0.3595 values between -0.35 and -0.23291 values between -0.23 and -0.111488 values between -0.11 and 0.021040 values between 0.02 and 0.14468 values between 0.14 and 0.27150 values between 0.27 and 0.3989 values between 0.39 and 0.5146 values between 0.51 and 0.6419 values between 0.64 and 0.7629 values between 0.76 and 0.8911 values between 0.89 and 1.014 values between 1.01 and 1.1310 values between 1.13 and 1.262 values between 1.26 and 1.381 values between 1.38 and 1.511 values between 1.51 and 1.631 values between 1.63 and 1.752 values between 1.75 and 1.882 values between 1.88 and 2.00offnegativepositiveoften

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. htable
  2. ldata
  3. eling
  4. imaginable
  5. 他会
  6. 蛉
  7. Ể
  8. 还会
  9. ورة
  10. efd

Suppresses

  1. 主张
  2. ชนะ
  3. mayoría
  4. 大多数
  5. favored
  6. reward
  7. 奖励
  8. 没收
  9. 多数
  10. ibu