Neuron Explorer

Qwen3-0.6B

pronouns referring back to a noun

A neuron in layer 10 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 responds to pronouns like "it," "one," "her" or "they" that point back to a thing or person mentioned earlier in the sentence. Its opposite side goes the other way, on he and they inside quoted speech. Turning it up didn't produce a clear theme in the sample text.

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

    In some people, however, cells start to ignore insulin, which means the pancreas must manufacture more of it. Insulin resistance is ‘silent’ because often the pancreas can keep up

  2. #2 peak

    able error message is received, it is in response to an HTTP request that includes a negation-related header such as Accept or Accept-Language. When the client includes one or more of these headers, the server will satisfy the requirements if possible.⏎The

  3. #3 peak

    we will do everything in our power to earn the confidence,” says Sten Roald Lorentzen.⏎The progress plan is not<|endoftext|>As process nodes shrink, it’s become increasingly difficult for the major semiconductor foundries to offer compelling advantages at each

  4. #4 peak

    ’t subjected to a long delay if a “cron job” happens to be running. Well almost…⏎Recently Dokuwiki gained an indexer to improve the performance of it’s search functionality. Dokuwiki doesn’t use a database – wiki pages are

  5. #5 peak

    and built a number of hiding places in the house and in the barn.⏎Galina and Yuri would bring food to Paulina each day. The family also provided her with clothing and false identity documents, should the farm be raided by the Germans.⏎

  6. #6 peak

    they’re treated:⏎Benign Paroxysmal Positional Vertigo (BPPV)⏎When someone has vertigo—a severe form of dizziness in which it feels like you or everything around

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 to be in the same group, and I have to be part of the group. The other group is also in the same group, but I have to be in the same group. So, the other group is in the same

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

    or an underlying set of principles, or an underlying way of looking at the world so that then when the next Aristotle comes around?" Jobs said. "Maybe if he carries around one these machines with him his whole life and types in all this stuff

  2. #2 peak

    the tech,” Ratto told Fast Company in an interview. “In fact you can’t even differentiate, you can’t even separate the tech from the social. They are irretrievably interwoven together. And I think this is the biggest

  3. #3 peak

    .⏎"If they were to replace the batteries with these supercapacitors, you could charge your

  4. #4 peak

    song; he talks about being pro-choice in this song, he talks about being anti-street-harassment in this song. But he also — it's an autobiographical song about being a young black male growing up<|endoftext|>Despite progress,

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. L7:1064 excites
  2. L9:175 suppresses
  3. L7:1405 excites
  4. L6:84 excites
  5. L7:1809 suppresses
  6. L8:1825 suppresses
  7. L9:1907 excites
  8. L5:583 excites
  9. L2:803 excites
  10. L5:165 suppresses
  11. L7:1624 excites
  12. L7:616 excites
  13. L6:1458 suppresses
  14. L6:1135 weakly suppresses
  15. L4:1306 weakly excites
  16. L9:592 weakly suppresses
  17. L7:173 weakly excites
  18. L2:2195 weakly suppresses
  19. L8:40 weakly suppresses
  20. L6:1191 weakly suppresses
  21. L9:52 weakly suppresses
  22. L5:1664 weakly suppresses
  23. L6:98 weakly excites
  24. L6:153 weakly excites

Feeds later layers

  1. L14:1668 suppresses
  2. L11:1419 excites
  3. L11:2864 excites
  4. L19:685 suppresses
  5. L14:3063 excites
  6. L15:396 suppresses
  7. L16:1077 weakly excites
  8. L12:1653 weakly excites
  9. L17:354 weakly excites
  10. L12:978 weakly excites
  11. L14:1317 weakly excites
  12. L11:222 weakly excites
  13. L15:503 weakly excites
  14. L14:990 weakly excites
  15. L14:1849 weakly excites
  16. L14:3022 weakly suppresses
  17. L18:381 weakly excites
  18. L15:1544 weakly excites
  19. L14:778 weakly suppresses
  20. L12:2347 weakly excites
  21. L17:202 weakly suppresses
  22. L18:2324 weakly excites
  23. L15:2811 weakly excites
  24. L12:850 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.

2 values between -1.90 and -1.761 values between -1.76 and -1.621 values between -1.48 and -1.344 values between -1.21 and -1.076 values between -1.07 and -0.931 values between -0.93 and -0.792 values between -0.65 and -0.5123 values between -0.51 and -0.37152 values between -0.37 and -0.24719 values between -0.24 and -0.101839 values between -0.10 and 0.04804 values between 0.04 and 0.18209 values between 0.18 and 0.3263 values between 0.32 and 0.4628 values between 0.46 and 0.608 values between 0.60 and 0.747 values between 0.74 and 0.876 values between 0.87 and 1.017 values between 1.01 and 1.155 values between 1.15 and 1.294 values between 1.29 and 1.435 values between 1.43 and 1.572 values between 1.57 and 1.712 values between 1.71 and 1.842 values between 1.84 and 1.982 values between 1.98 and 2.121 values between 2.26 and 2.401 values between 2.54 and 2.681 values between 3.09 and 3.231 values between 3.51 and 3.65offnegativepositiveoften

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. erves
  2. 个工作日
  3. erved
  4. ở
  5. den
  6. <header
  7. EXTERN
  8. _DIRECT
  9. 主持召开
  10. 服务能力

Suppresses

  1. mieux
  2. 不上
  3. 高低
  4. Karnataka
  5. jang
  6. loff
  7. ��
  8. ogne
  9. 但他们
  10. stair