Neuron Explorer

Qwen3-0.6B

spatial prepositions after motion verbs

A neuron in layer 16 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 the small word that tells where something is positioned or moving, such as 'through', 'around', 'inside', 'against' or 'with', right after a verb of insertion, wrapping or contact. Turning it up made the model write unrelated text, so the examples are the main evidence.

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

    . This minor procedure uses a laparoscope, a viewing instrument with a light, to see your uterus and pelvis. The laparoscopy is inserted through a small incision in your belly. This procedure is generally performed under general anesthesia

  2. #2 peak

    the uterus to create a more transparent image. Hysteroscopy is another common test that involves inserting a tiny telescope through the vagina and cervix. Lastly, the doctor may choose to perform an end

  3. #3 peak

    part of getting your belly healthy again. As you can see in the image above, the TA is a thick layer of muscle. The transverse abdominal muscle wraps around the torso from front to back and the muscle fibers of the TA run horizontally,

  4. #4 peak

    is a cancer of the prostate gland. The prostate gland is a walnut-sized gland present only in men, found in the pelvis below the bladder. It wraps around the urethra (the tube that extends from the bladder) and lies in

  5. #5 peak

    -volvus – condition in which the stomach dilates with gas and/or fluid, and subsequently rotates around its short axis⏎- Heart failure⏎- Heartworm disease⏎- Heat stroke

  6. #6 peak

    A Comparison Of High-Level And Low-Level Programming Languages⏎A way to categorize programming languages is through high-level or low-level. The low-level languages communicate directly with the central processing unit or the CPU. They can perform very basic commands but are

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 going to my own home, or my own home as I live my life as I live my life as I live my life as I live my life as I live my life as I live my life as I live my life

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

    to go into the world and play an active role in its creation and recreation. This seems to me an endeavor fundamental to the human condition.⏎Where I diverge from Zhao and Robinson is in their discussion of creating an entrepreneurial mindset.⏎Because entrepreneurs bring

  2. #2 peak

    of antidepressant medication and psychotherapy is the best prescription.⏎One very common misconception about depression is that it feels like mere sadness. Sadness is only a part of it. It

  3. #3 peak

    The advantages and applications of this approach are manifold. Not only can it recover low-grade heat for power generation, it can also be done using thinner, lighter devices than semi-conductors. One slim cell is all that is needed to regulate the

  4. #4 peak

    that are imported from the procurement supply warehouse or purchased locally. This type of warehouse is generally close to the commodity sales market, and the scale is generally smaller than that of the procurement supply warehouse. Also engaged in the zero-removal supply

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. L13:68 excites
  2. L13:2465 excites
  3. L15:942 suppresses
  4. L15:1634 excites
  5. L9:2806 suppresses
  6. L12:2833 excites
  7. L10:132 excites
  8. L11:2207 suppresses "of" after a noun in trademark and title phrases
  9. L8:1003 suppresses "the" or "of" before a noun phrase
  10. L15:1053 excites
  11. L15:2613 suppresses
  12. L15:2012 suppresses
  13. L6:1637 excites
  14. L12:850 excites
  15. L11:1910 excites
  16. L13:1054 excites
  17. L13:1173 excites
  18. L14:989 suppresses
  19. L15:2740 weakly suppresses
  20. L15:1227 weakly suppresses
  21. L15:2231 weakly excites
  22. L11:1174 weakly suppresses
  23. L15:894 weakly excites
  24. L15:1741 weakly suppresses

Feeds later layers

  1. L17:2267 excites
  2. L18:2071 weakly excites
  3. L17:2199 weakly excites
  4. L17:2753 weakly excites
  5. L18:1918 weakly excites
  6. L18:2321 weakly excites
  7. L20:1910 weakly excites
  8. L18:1143 weakly excites
  9. L19:241 weakly suppresses
  10. L18:381 weakly suppresses
  11. L18:1860 weakly excites
  12. L21:645 weakly excites
  13. L20:611 weakly suppresses
  14. L20:358 weakly suppresses
  15. L18:183 weakly suppresses
  16. L17:2281 weakly suppresses
  17. L17:140 weakly suppresses
  18. L17:182 weakly suppresses
  19. L20:554 weakly excites
  20. L21:599 weakly excites
  21. L19:650 weakly excites
  22. L21:1160 weakly excites 'is' after a subject noun about likelihood
  23. L22:596 weakly excites
  24. L18:2669 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 -6.60 and -6.293 values between -4.10 and -3.792 values between -3.79 and -3.473 values between -3.47 and -3.163 values between -3.16 and -2.851 values between -2.85 and -2.533 values between -2.53 and -2.2212 values between -2.22 and -1.9111 values between -1.91 and -1.5915 values between -1.59 and -1.2830 values between -1.28 and -0.9749 values between -0.97 and -0.66119 values between -0.66 and -0.34830 values between -0.34 and -0.031520 values between -0.03 and 0.28571 values between 0.28 and 0.60302 values between 0.60 and 0.91152 values between 0.91 and 1.2297 values between 1.22 and 1.5469 values between 1.54 and 1.8535 values between 1.85 and 2.1628 values between 2.16 and 2.4823 values between 2.48 and 2.7912 values between 2.79 and 3.106 values between 3.10 and 3.411 values between 3.41 and 3.732 values between 3.73 and 4.043 values between 4.04 and 4.351 values between 4.67 and 4.982 values between 5.29 and 5.612 values between 5.61 and 5.92offnegativepositiveoften

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. ocoder
  2. тай
  3. (parts
  4. iosis
  5. parts
  6. 美德
  7. Parts
  8. 零件
  9. alus
  10. abies

Suppresses

  1. führer
  2. стр
  3. RIES
  4. 作为一个
  5. 学堂
  6. IRTUAL
  7. 书房
  8. UGH
  9. 通用
  10. 、「