Neuron Explorer

Qwen3-0.6B

possessive pronouns for a group

A neuron in layer 8 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 words like "their", "his" and "they" when they point back to a group of people mentioned earlier, such as "the hours they spent". Its strongest negative reactions come on "who" starting a clause about people. Turning it up gave unrelated text, so steering does not confirm the pattern.

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

    His voice⏎Jeremiah 7:28 "And you shall say to them, 'This is the nation that did not obey the voice of the LORD their God or accept correction; truth has perished and has been cut

  2. #2 peak

    in the harm principle:⏎That principle is, that the sole end for which mankind are warranted, individually or collectively in interfering with the liberty of action of any of their number, is self-protection. That the only purpose for which power can be

  3. #3 peak

    . This data, which used nationally representative samples, were collected just prior to the release of ChatGPT. The countries we surveyed are leaders in AI activity within their regions, as evidenced by their investment in AI and AI-specific employment.⏎Do employees

  4. #4 peak

    been asking their coaches to add gardening and landscaping activities into their routines,” Mandal said, “because

  5. #5 peak

    Skill practices and on-the-spot defensive driving techniques help change bravado to confidence.⏎DDC-Alive at 25 teaches young adults that:⏎* People in their age group are more likely to be hurt or killed in a vehicle crash.⏎*

  6. #6 peak

    of his trips, he so impressed the bishop of Cesarea that he was ordained. However, his writings, not quite orthodox, were questioned. A bishop of his acquaintance ordered him excommunicated. Pontian was required to hold a synod

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 will be something else." What is the best for the next month? Oh, that's the same as the previous one. So, "I will be something else." What is the best for the next month? I

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

    bottom arm extended beneath shoulder (or forearm supporting body), other arm extended upward and legs crossed top over bottom. Partner 2 stands slightly behind Partner 1, who lifts hip while reaching under body with top arm, then returns top arm to extended

  2. #2 peak

    would have profaned the mistletoe to fall upon the ground. He would then divide the branches into many sprigs and distributed them to the people, who hung them over doorways as protection against thunder, lightning and other evils.⏎The

  3. #3 peak

    a lot of other versions that appear to derive from the Acts of Thomas version.⏎Thomas, traveling with a merchant into India, was brought before Gundaphorus, who asked him his craft. Thomas said that he was a carpenter and a builder

  4. #4 peak

    Mountain Light, wrote a message to photographers written from the point of view of a roll of film: “The vision through your eyes is filtered through your brain, which interprets colors for you before you ‘see’ them. Humans must have had

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. L7:2482 suppresses
  2. L6:549 suppresses
  3. L4:1306 excites
  4. L4:388 weakly suppresses
  5. L2:2981 weakly suppresses
  6. L4:885 weakly excites
  7. L2:2903 weakly excites
  8. L6:1653 weakly excites
  9. L7:312 weakly excites
  10. L4:1282 weakly suppresses
  11. L0:1363 weakly suppresses
  12. L0:1402 weakly excites
  13. L1:283 weakly suppresses
  14. L1:2096 weakly excites
  15. L6:2067 weakly excites
  16. L6:1184 weakly suppresses
  17. L3:2570 weakly suppresses
  18. L3:551 weakly excites
  19. L0:1926 weakly excites
  20. L3:1340 weakly excites
  21. L4:1817 weakly suppresses
  22. L4:2211 weakly excites
  23. L1:242 weakly excites
  24. L7:2220 weakly suppresses

Feeds later layers

  1. L16:1072 suppresses
  2. L9:172 weakly suppresses
  3. L13:605 weakly excites
  4. L12:559 weakly suppresses
  5. L12:981 weakly suppresses
  6. L10:959 weakly suppresses
  7. L17:124 weakly excites
  8. L16:2276 weakly suppresses
  9. L9:2816 weakly suppresses
  10. L20:72 weakly suppresses
  11. L19:120 weakly excites
  12. L20:169 weakly suppresses
  13. L20:188 weakly suppresses
  14. L15:2521 weakly excites
  15. L14:168 weakly suppresses
  16. L10:430 weakly suppresses
  17. L20:190 weakly suppresses
  18. L10:2751 weakly suppresses
  19. L21:2609 weakly excites
  20. L9:1907 weakly suppresses
  21. L17:448 weakly suppresses
  22. L9:120 weakly excites
  23. L18:715 weakly suppresses
  24. L17:112 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.

2 values between -5.64 and -5.421 values between -3.25 and -3.032 values between -2.81 and -2.602 values between -2.16 and -1.943 values between -1.94 and -1.724 values between -1.72 and -1.5110 values between -1.51 and -1.296 values between -1.29 and -1.0712 values between -1.07 and -0.8527 values between -0.85 and -0.6446 values between -0.64 and -0.42221 values between -0.42 and -0.201919 values between -0.20 and 0.021355 values between 0.02 and 0.23192 values between 0.23 and 0.4544 values between 0.45 and 0.6723 values between 0.67 and 0.899 values between 0.89 and 1.1013 values between 1.10 and 1.326 values between 1.32 and 1.543 values between 1.54 and 1.763 values between 1.76 and 1.983 values between 1.98 and 2.191 values between 2.41 and 2.631 values between 2.85 and 3.06offnegativepositiveoften

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. orc
  2. яз
  3. UserController
  4. ORIA
  5. timeofday
  6. >manual
  7. hör
  8. interviewed
  9. licit
  10. lying

Suppresses

  1. uture
  2. leaf
  3. ern
  4. unities
  5. ages
  6. uri
  7. uropean
  8. eny
  9. si
  10. annt