Neuron Explorer

Qwen3-0.6B

unclear: pronouns in advice and history text

A neuron in layer 12 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 responds to pronouns and name-like word pieces in mixed text. It is strongest in a few spots and strongly negative in parenting advice, where it reacts to words like 'who', 'you' and 'your' referring to parents and children. Turning it up made the model drift to unrelated topics.

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

    Holocaust and of willful radical evil in our time.⏎What the liberators found in these camps astounded them. They came as soldiers, and left as liberators. They

  2. #2 peak

    Defines the area covered by the Treaty as everything south of 60˚S latitude.⏎Article VII: Observers may be designated by Contracting Parties to carry out inspections of stations, equipment, and expeditions, and will

  3. #3 peak

    that activities contrary to the spirit of the Treaty do not occur.⏎Article XI: Disputes between Contracting Parties should be resolved peacefully.<|endoftext|>Read: The Heartbeat of Trees: Em

  4. #4 peak

    highlight the impact of the uneven distribution of responsibilities on the leader and on the group. Ask members about the division of responsibilities in their groups. Ask some leaders how they feel about this.⏎Brief participants to prepare for a role-play, either using some

  5. #5 peak

    southwestern United States. So how can National Geographic say that today’s civilizations are at fault?⏎“The medieval Khmer were confronted with a period of climatic instability that they had no experience of, and which fully changed the rules of the game that they

  6. #6 peak

    <|endoftext|>On a September evening in 1931, John and Jack, two of the caretakers of the Imaginarium Geographica, discover a plea for help on a medieval manuscript - which seems to have been written

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 am, for example, a person who is, in the beginning of the school, when the school is on the hill, when the school is on the hill, when the school is on the hill, when the school is on

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

    anxiety. You want to teach your kids how to cope, not just simply remove them from all stressful situations. It is best to lead by example: a parent who addresses their anxieties and calmly overcomes will teach their children to do the same

  2. #2 peak

    2-year-old how to practice generosity — both within your family and away from home.⏎How to teach your child honesty⏎Start early to show your child that you trust him and that honesty is important to you, and you'll lay the groundwork

  3. #3 peak

    of your child. Be a role model for them. Kids are the best observers, and they tend to imitate there parents.⏎- Show the various options to your child, discuss books with them, and more importantly allow them to choose what they

  4. #4 peak

    , and close with a positive. Maybe the best advice is to comment in the style and mood that you would like to receive.⏎When you comment, show that you are listening. Respond to specific parts of what the author wrote. As

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. L9:2746 excites
  2. L8:1868 excites
  3. L10:2757 excites
  4. L11:422 suppresses
  5. L11:2071 suppresses
  6. L11:1617 excites
  7. L11:2337 excites
  8. L11:761 excites
  9. L10:2620 excites
  10. L7:505 excites
  11. L8:872 excites
  12. L4:1562 excites
  13. L11:2023 suppresses
  14. L11:1704 suppresses
  15. L11:1108 excites
  16. L11:2754 excites
  17. L9:652 weakly suppresses
  18. L6:195 weakly suppresses
  19. L10:153 weakly suppresses
  20. L8:246 weakly excites
  21. L6:2239 weakly suppresses
  22. L10:2196 weakly suppresses
  23. L9:2679 weakly suppresses
  24. L8:2733 weakly suppresses

Feeds later layers

  1. L18:2131 suppresses
  2. L14:2012 excites
  3. L19:2460 excites
  4. L13:1608 suppresses
  5. L15:3067 excites
  6. L13:789 excites
  7. L18:2606 suppresses
  8. L15:1815 excites
  9. L13:1448 excites
  10. L13:1879 suppresses
  11. L14:2178 suppresses
  12. L19:2191 suppresses
  13. L13:1697 excites
  14. L17:1893 suppresses
  15. L14:1845 suppresses
  16. L13:2151 suppresses
  17. L21:1473 weakly suppresses
  18. L13:1639 weakly suppresses
  19. L13:1514 weakly suppresses
  20. L14:1529 weakly suppresses
  21. L25:2480 weakly excites
  22. L15:1643 weakly suppresses
  23. L15:1461 weakly suppresses
  24. L20:1936 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 -7.08 and -6.851 values between -2.88 and -2.651 values between -2.65 and -2.422 values between -2.19 and -1.959 values between -1.95 and -1.723 values between -1.72 and -1.4912 values between -1.49 and -1.2523 values between -1.25 and -1.0231 values between -1.02 and -0.7953 values between -0.79 and -0.55161 values between -0.55 and -0.32505 values between -0.32 and -0.092374 values between -0.09 and 0.15417 values between 0.15 and 0.38165 values between 0.38 and 0.6167 values between 0.61 and 0.8536 values between 0.85 and 1.0823 values between 1.08 and 1.3111 values between 1.31 and 1.556 values between 1.55 and 1.783 values between 1.78 and 2.014 values between 2.01 and 2.25offnegativepositiveoften

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. automation
  2. 旅客
  3. führung
  4. azar
  5. cott
  6. instead
  7. 鸡汤
  8. 来宾
  9. EFI
  10. USAGE

Suppresses

  1. Own
  2. own
  3. <![
  4. uisse
  5. ��
  6. ythe
  7. Público
  8. own
  9. itations
  10. ző