Neuron Explorer

Qwen3-0.6B

colon after a quoted source line

A neuron in layer 9 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 lights up on the colon in the word 'Source:' that comes right after a quotation, the kind of line that credits where a passage came from. It is most active in web text that ends with a citation. When it was turned up, the model's writing drifted into unrelated text and quiz-style answers, so the steering did not show a clear theme.

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

    , atomic force microscopes can easily be outfitted for detailed relative humidity and reactive gas studies to further probe processes of chemical importance under environmentally applicable conditions.<|endoftext|>Source: World Health Organization⏎In 2015, antiretroviral

  2. #2 peak

    not relieve symptoms once they occur but do prevent the late-phase response. All patients with chronic asthma require anti-inflammatory drugs to control their response to allergens.⏎Source: Mayo Clinic, Heal Asthma

  3. #3 peak

    active components of DST can be replicated in other school environments.⏎Source: Health Behavior and Policy Review, Volume 8, Number 4, July

  4. #4 peak

    was a “bride,” so that one can easily tell at a glance if a man be or be not mortgaged as to his affections.⏎Source: "German Betrothal Customs," (Harvard) Herald, 6 October

  5. #5 peak

    sprouts it grows and suddenly bears a poison fruit which we must pluck.’⏎Source: John Keats, 14 Feb – 3 May 181

  6. #6 peak

    still exists. We need to ensure there is broad and consistent application of proven tobacco cessation interventions if we are to sustain the overall progress documented in this study.”⏎Source: American Cancer Society⏎Explore further: AMA: avoiding distress in medical school<|endoftext|>Be

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 a man who has been given a life of freedom. But I am not the only one. How can I feel confident? I am a woman with a right to freedom. I am a woman with a right to life. 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

    in same size⏎C. No required of subnet⏎D. All of above⏎6-3. What does the port number in a TCP connection specify?⏎A. It specifies the communication process on the

  2. #2 peak

    & connection⏎C. It specify the size of data⏎D. All of the above⏎6-4. The class-based addressing is also known as⏎A. Modern Model⏎B. Classful Model⏎C. Classless Model⏎

  3. #3 peak

    / Objective Questions are given below of this question set.⏎Multiple Choice Questions of Computer Networking⏎6-1. What is the IP Address range of APIPA?⏎A. 169.254.0.1 to 1

  4. #4 peak

    -7. Which of the following is reliable communication?⏎D. All of them⏎6-8. What is the typical range of Ephemeral ports?⏎A. 1 to 80⏎B. 1 to 10

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. L7:64 suppresses
  2. L7:1492 weakly suppresses
  3. L8:341 weakly excites
  4. L3:147 weakly suppresses
  5. L5:233 weakly suppresses
  6. L8:7 weakly excites
  7. L8:355 weakly suppresses
  8. L7:573 weakly excites
  9. L7:1574 weakly excites
  10. L6:18 weakly excites
  11. L4:568 weakly excites
  12. L2:639 weakly excites
  13. L1:923 weakly suppresses
  14. L8:10 weakly excites
  15. L3:187 weakly excites
  16. L4:2250 weakly excites
  17. L7:28 weakly excites
  18. L3:66 weakly suppresses
  19. L5:1974 weakly excites
  20. L5:1621 weakly suppresses
  21. L1:2662 weakly suppresses
  22. L6:2464 weakly excites
  23. L7:563 weakly excites
  24. L0:537 weakly excites the 'or' in 'and/or'

Feeds later layers

  1. L15:1624 suppresses
  2. L19:2294 excites
  3. L14:2119 weakly excites
  4. L18:1943 weakly excites
  5. L20:1763 weakly suppresses
  6. L10:1472 weakly suppresses
  7. L11:2616 weakly excites
  8. L11:2934 weakly excites
  9. L13:2884 weakly excites
  10. L19:1271 weakly suppresses
  11. L12:1893 weakly excites
  12. L23:1323 weakly excites
  13. L14:2663 weakly suppresses
  14. L15:1906 weakly suppresses
  15. L13:834 weakly suppresses
  16. L16:1981 weakly excites
  17. L22:2524 weakly suppresses
  18. L26:1656 weakly suppresses
  19. L13:2242 weakly suppresses
  20. L22:347 weakly suppresses
  21. L18:1802 weakly suppresses
  22. L17:1056 weakly excites
  23. L14:69 weakly excites
  24. L14:2345 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 -1.08 and -1.021 values between -0.79 and -0.732 values between -0.67 and -0.622 values between -0.62 and -0.566 values between -0.56 and -0.5014 values between -0.50 and -0.4436 values between -0.44 and -0.3964 values between -0.39 and -0.33105 values between -0.33 and -0.27174 values between -0.27 and -0.21259 values between -0.21 and -0.16435 values between -0.16 and -0.10553 values between -0.10 and -0.04710 values between -0.04 and 0.02594 values between 0.02 and 0.07422 values between 0.07 and 0.13235 values between 0.13 and 0.19164 values between 0.19 and 0.2570 values between 0.25 and 0.3023 values between 0.30 and 0.3622 values between 0.36 and 0.428 values between 0.42 and 0.482 values between 0.48 and 0.531 values between 0.53 and 0.591 values between 0.59 and 0.651 values between 0.71 and 0.761 values between 0.76 and 0.821 values between 0.82 and 0.881 values between 1.17 and 1.23offnegativepositiveoften

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. DataTask
  2. outings
  3. ophage
  4. eman
  5. yük
  6. entifier
  7. culos
  8. 年人
  9. autoload
  10. もら

Suppresses

  1. forwards
  2. pred
  3. 富裕
  4. Free
  5. material
  6. 内的
  7. arg
  8. ち
  9. 钢材
  10. forward