Neuron Explorer

Qwen3-0.6B

comma closing 'the more X, the more Y' clauses

A neuron in layer 21 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 at the end of a 'the more you do this' phrase, just before the second half of the sentence that starts with 'the more'. Turning it down made the model get stuck repeating the word 'more'.

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

    few (or 30) seconds or your water’s a couple degrees off temperature, you’re NOT going to ruin your film. The more accurate you are,

  2. #2 peak

    this is done, the company loses vital manufacturing and quality information. This information accurately collected is the only basis of implementing quality improvement. The more granular the information, the more effective it can become.⏎If the only options are in-tolerance and

  3. #3 peak

    spinning shaft. The propeller is attached to that shaft. At least two (but usually three or four) blades make up the propeller. The more blades, the more air that can be moved by the propeller. A blade has an

  4. #4 peak

    are being used to validate this identity combination, and to prove a virtual person is indeed the same person in real life. The more value an identity has access to, more effort and work is required to validate and establish secure authentications for it.⏎

  5. #5 peak

    stress.⏎Stress is an underlying factor in illness as it reduces your immune system’s ability to fight off antigens. This means that the more stressed you are, the likelier you are to pick up infections and fight illness.⏎Therefore, a

  6. #6 peak

    , to a jog in the 20th century, to a sprint now.⏎The reason for this is simple. The more technology you have at your disposal, the easier it is to invent new things. Imagine trying to build a computer using

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: more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more more

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

    played by the unpredictable in war,” Thucydides wrote, “and think of it now, before you are actually committed. The longer a war lasts, the more things tend to depend on accidents. Neither you nor we can see into them

  2. #2 peak

    that there are 10 billion trillion stars in the Universe, but the fact is, they will never know for sure. The deeper into space we look, the more stars we fi nd. How remarkable it is to think our

  3. #3 peak

    it important that it be crowned and formally installed at the Shrine of Christ the King in Chicago, IL by Cardinal George? A. "The more you honor Me the more I will bless you" is the promise of the Infant King. The coron

  4. #4 peak

    once you see the incredible benefits you could never look back. It cooks your food and keeps it piping hot. The more you incorporate it into your daily life, the more you will find you use it.⏎A degree of experimentation is necessary to get

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. L18:3019 strongly suppresses
  2. L20:2321 strongly excites
  3. L20:2323 excites
  4. L20:944 excites
  5. L19:1430 excites 'the more X' clause before 'the more'
  6. L19:2597 excites
  7. L19:2024 excites
  8. L20:2852 excites
  9. L20:2580 excites
  10. L20:2477 suppresses
  11. L18:2761 excites
  12. L20:2491 suppresses
  13. L17:2344 suppresses
  14. L18:2316 excites
  15. L18:2875 excites
  16. L20:885 suppresses 'and' joining paired words
  17. L20:2564 weakly excites
  18. L17:2278 weakly excites
  19. L18:824 weakly suppresses
  20. L19:2747 weakly excites
  21. L9:900 weakly excites
  22. L18:2814 weakly excites
  23. L19:196 weakly suppresses
  24. L18:2935 weakly excites

Feeds later layers

  1. L26:551 strongly suppresses
  2. L24:847 strongly suppresses
  3. L27:948 strongly suppresses
  4. L27:1896 strongly suppresses split names and comparative stems
  5. L22:1630 suppresses
  6. L27:1734 suppresses
  7. L26:1908 suppresses
  8. L23:456 suppresses
  9. L25:2908 excites
  10. L24:2350 suppresses
  11. L22:1988 suppresses "the" in "the more..., the more" phrases
  12. L22:1700 suppresses
  13. L25:1046 suppresses
  14. L25:1196 suppresses
  15. L23:3034 weakly suppresses
  16. L26:124 weakly excites
  17. L23:380 weakly suppresses
  18. L25:884 weakly suppresses
  19. L27:2471 weakly suppresses
  20. L23:875 weakly suppresses
  21. L25:461 weakly suppresses
  22. L22:872 weakly excites
  23. L23:939 weakly suppresses
  24. L24:1488 weakly 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 -25.38 and -24.631 values between -22.40 and -21.651 values between -19.42 and -18.681 values between -18.68 and -17.932 values between -10.49 and -9.751 values between -9.00 and -8.263 values between -8.26 and -7.521 values between -7.52 and -6.771 values between -6.77 and -6.031 values between -6.03 and -5.284 values between -5.28 and -4.541 values between -4.54 and -3.801 values between -3.80 and -3.057 values between -3.05 and -2.3112 values between -2.31 and -1.5634 values between -1.56 and -0.82516 values between -0.82 and -0.072889 values between -0.07 and 0.67305 values between 0.67 and 1.41112 values between 1.41 and 2.1611 values between 2.16 and 2.902 values between 2.90 and 3.651 values between 3.65 and 4.39offnegativepositiveoften

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. iesen
  2. .patch
  3. modo
  4. arend
  5. ילת
  6. ulg
  7. led
  8. .inject
  9. ulous
  10. regation

Suppresses

  1. more
  2. 更
  3. 更多
  4. 更多的
  5. more
  6. 更
  7. 更有
  8. 更为
  9. More
  10. 更强