Neuron Explorer

Qwen3-0.6B

semicolon joining clauses

A neuron in layer 18 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 semicolons that join one clause to a second, explaining part, in long sentences of prose. When it was turned up, the model's writing filled up with semicolons.

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

    pulling together genetic, historical and sociocultural factors. The paper aimed to explain why African Americans are more likely to abstain or drink less compared to European Americans; why those who do drink encounter more negative consequences; and which African American population is

  2. #2 peak

    . This aggravates some persisting challenges, such as the large number of unqualified preschool teachers or teachers that have not been trained to teach at the preschool level; curricula and pedagogies that are not adaptable to the various developmental stages of

  3. #3 peak

    commonly known as the draft. For those who are not aware, the draft is a system of conscription that enlists citizens into the military; the system is run using an online sign up form that assigns you a draft number

  4. #4 peak

    of the mouth – one on each side of the mouth. These small organs help the immune system trap bacteria, viruses, and other harmful toxins that we inhale; preventing illness. Though they perform such a valuable task in helping the body remain healthy

  5. #5 peak

    and inter-disciplinary efforts to create a sustainable society. In fact, achieving the goal of environmental, economic, and social sustainability is predicated on changing human behavior; the purview of Psychologists (reviewed in , see also [2,

  6. #6 peak

    Housing Act of 1954, however, introduced federal funding for rehabilitation-based approaches as well. This talk considers the motivations behind this more conservationist approach; the practical constraints to its wider-spread adoption; and its prevalence, character,

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. I, I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. I. 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

    angular velocity is ω;⏎③ The blade rotates around its geometric center, and the angular velocity is ω 0, and the velocity is ω; on the premise that the cutter position q is equal to the blade radius r,

  2. #2 peak

    ⏎t0=0; %set the initial time to zero⏎tend=1; %set the final time to 100⏎tspan=[t0

  3. #3 peak

    research brief focuses on the following questions: 1) How does time in spent in foster care affect a child’s chances of continuing to live in foster care; and 2) Is child age at the time of a maltreatment investigation associated with

  4. #4 peak

    achea (in male striped seals) or from the esophagus (in male walruses) and serve to alter the specific gravity of the body; and (4) in the sperm whale, a blind air sac about 1 cu

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. L17:579 strongly excites semicolons joining two clauses
  2. L17:1069 suppresses
  3. L17:2027 suppresses
  4. L14:155 excites sentence-ending periods in prose
  5. L16:1078 suppresses
  6. L5:5 weakly suppresses digit after a label like paper or type
  7. L4:51 weakly excites
  8. L9:1675 weakly excites
  9. L17:1772 weakly suppresses
  10. L16:1130 weakly suppresses
  11. L9:819 weakly suppresses
  12. L17:519 weakly suppresses
  13. L16:2817 weakly excites
  14. L10:33 weakly suppresses
  15. L15:286 weakly excites
  16. L17:990 weakly excites
  17. L15:1473 weakly suppresses
  18. L12:2487 weakly suppresses
  19. L14:2819 weakly excites
  20. L13:2476 weakly suppresses
  21. L8:76 weakly suppresses
  22. L0:1488 weakly suppresses
  23. L10:2615 weakly suppresses
  24. L16:1040 weakly excites

Feeds later layers

  1. L27:347 strongly excites
  2. L26:421 strongly excites
  3. L21:351 strongly excites
  4. L20:373 strongly excites
  5. L23:171 suppresses
  6. L26:603 excites
  7. L20:206 suppresses
  8. L24:531 excites
  9. L24:378 excites
  10. L25:238 excites
  11. L27:958 excites
  12. L21:759 suppresses
  13. L25:318 excites
  14. L25:296 excites
  15. L21:233 suppresses
  16. L19:1130 suppresses
  17. L24:264 excites
  18. L23:100 suppresses comma or full stop ending a clause
  19. L23:201 excites
  20. L25:542 excites
  21. L25:297 suppresses
  22. L25:40 excites
  23. L20:637 suppresses
  24. L27:207 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.

1 values between -5.92 and -5.152 values between -5.15 and -4.374 values between -4.37 and -3.6010 values between -3.60 and -2.8327 values between -2.83 and -2.0540 values between -2.05 and -1.28103 values between -1.28 and -0.513465 values between -0.51 and 0.27216 values between 0.27 and 1.048 values between 1.04 and 1.8111 values between 1.81 and 2.586 values between 2.58 and 3.362 values between 3.36 and 4.131 values between 4.13 and 4.901 values between 4.90 and 5.681 values between 8.00 and 8.772 values between 8.77 and 9.542 values between 11.86 and 12.631 values between 12.63 and 13.411 values between 14.95 and 15.731 values between 16.50 and 17.271 values between 22.68 and 23.452 values between 24.23 and 25.00offnegativepositiveoften

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. ;
  2. ”;
  3. ;
  4. ";
  5. ';
  6. ;");⏎
  7. ;";⏎
  8. ();
  9. ;<
  10. ;"

Suppresses

  1. ⏎⏎
  2. ⏎ ⏎
  3. **⏎⏎
  4. **⏎
  5. ⏎ ⏎
  6. ⏎⏎⏎
  7. }⏎⏎
  8. ⏎
  9. ⏎ ⏎
  10. }⏎⏎