Neuron Explorer

Qwen3-0.6B

comma after 'For X,' introducing a viewpoint

A neuron in layer 14 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 follows a phrase like "For an amputee," or "For dairy producers,", where the writer is about to say how something looks or feels from that person's point of view. Steering it did not produce text that clearly reflected this, so the link is only seen in the examples.

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

    of the most common questions I get is, “Why do we separate the calves from their mums?”⏎These questions invariably come from mothers. For dairy producers, this is so much a part of

  2. #2 peak

    and universities across the country.⏎In Pakistan, schools, colleges, and universities are considered to be safe spaces for learning and personal growth. However, for many students, these institutions can become a source of fear, anxiety

  3. #3 peak

    their cute googly eyes.⏎The world looks colorful to our eyes with only three color receptors — red, green, and blue. But for a mantis shrimp, a small marine crustacean with 12 color receptors, it looks quite

  4. #4 peak

    or a performance; it really did upset the general.⏎Then his long, tall shadow fell across the young Hoosier from northern Indiana. For Private Kuhl, it must have felt like

  5. #5 peak

    ⏎A new breakthrough in prosthesis has delivered a bionic hand that is capable of feeling just like the organic limb it replaces.⏎For an amputee, replacing a missing limb with a functional prosthetic can alleviate physical or emotional distress and

  6. #6 peak

    . Dance is one of the most human activities ever created and, luckily for us, it’s not going away any time soon. For Physie girls and women, dancing is like breathing, eating and sleeping… it’s a necessity!⏎One answer

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 the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the, the

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

    events,” he says. “That’s just what creationists say can’t happen.”’1⏎The many comments posted on the New Scientist website shows just how excited the atheists are about this report. They are positively gloating.⏎In 1

  2. #2 peak

    . MTBE has been detected about 4 to 6 times more frequently in the high-use areas than elsewhere.⏎In addition to its studies of ambient ground water, USGS is<|endoftext|>

  3. #3 peak

    ’s death on the Cross, descent into Hell, and the Resurrection.⏎Reproaching the lack of penitence and recalcitrance of the Jews, the Lord said, “The Ninevites shall rise in the judgment with this

  4. #4 peak

    ization as a World Church.”1 The Council marked the definitive end of European influence and domination of the Catholicity of the Church. With farsighted acumen, Rahner suggested that the process of creating an authentically polycentric World

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. L11:1835 excites
  2. L6:2326 suppresses
  3. L9:2070 suppresses
  4. L11:487 excites
  5. L9:1627 excites
  6. L13:1535 suppresses
  7. L13:2636 suppresses
  8. L12:1352 suppresses
  9. L12:1667 excites
  10. L9:1550 excites 'to' after 'matters' or 'means' phrases
  11. L8:1703 excites
  12. L7:194 excites
  13. L9:2202 weakly suppresses
  14. L6:1001 weakly suppresses
  15. L12:2347 weakly suppresses
  16. L13:287 weakly excites
  17. L5:247 weakly suppresses
  18. L13:2465 weakly suppresses
  19. L3:1956 weakly suppresses
  20. L13:1100 weakly suppresses
  21. L13:541 weakly excites
  22. L3:520 weakly suppresses
  23. L5:2987 weakly suppresses
  24. L6:1637 weakly excites

Feeds later layers

  1. L16:1219 strongly suppresses
  2. L15:2005 suppresses
  3. L16:1441 excites
  4. L20:862 suppresses
  5. L18:1629 suppresses
  6. L16:2208 suppresses
  7. L15:1919 excites
  8. L15:947 suppresses
  9. L16:1382 suppresses comma after an introductory phrase
  10. L18:2708 suppresses
  11. L15:1226 suppresses
  12. L18:1199 suppresses
  13. L15:942 excites
  14. L17:1471 suppresses
  15. L17:2546 excites
  16. L15:3044 suppresses
  17. L15:1771 suppresses
  18. L18:1576 suppresses
  19. L15:2633 suppresses
  20. L19:2508 suppresses
  21. L18:1584 suppresses
  22. L18:2195 suppresses
  23. L16:1998 suppresses
  24. L16:564 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 -6.79 and -6.581 values between -5.30 and -5.081 values between -4.87 and -4.661 values between -3.59 and -3.371 values between -3.16 and -2.954 values between -2.95 and -2.731 values between -2.52 and -2.303 values between -2.30 and -2.094 values between -2.09 and -1.885 values between -1.88 and -1.667 values between -1.66 and -1.4510 values between -1.45 and -1.2417 values between -1.24 and -1.0222 values between -1.02 and -0.8129 values between -0.81 and -0.5969 values between -0.59 and -0.38312 values between -0.38 and -0.171924 values between -0.17 and 0.051226 values between 0.05 and 0.26232 values between 0.26 and 0.4825 values between 0.48 and 0.699 values between 0.69 and 0.901 values between 0.90 and 1.122 values between 1.12 and 1.331 values between 1.54 and 1.76offnegativepositiveoften

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. ch
  2. Interr
  3. interruption
  4. enburg
  5. ky
  6. ime
  7. interceptions
  8. 打断
  9. 和服务
  10. ража

Suppresses

  1. scientists
  2. 科学家
  3. chàng
  4. ]: ⏎
  5. xFFFFFF
  6. Bosch
  7. aders
  8. 统计局
  9. */ ⏎
  10. exels