Neuron Explorer

Qwen3-0.6B

textbook and study-help wording

A neuron in layer 7 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 fires on words common in textbooks, homework help and edition listings, such as 'answers', 'Edition', 'data' and 'expert'. When turned up, the model's text often shifted toward textbook problems and worked examples.

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

    ids and Bases practice, you can also practice Conjugate Acids and Bases practice problems.⏎What professor is this problem relevant for?⏎Based on our data, we think this problem is relevant for Professor Finnegan's class at WSU

  2. #2 peak

    just stop by for help. By getting the help of a tutor, the student can better understand their coursework.⏎Check the Textbook⏎Many textbooks will actually contain answers in the back of the book. For the more difficult problems, the

  3. #3 peak

    keeping us alive.⏎A good day for the amygd<|endoftext|>Calculate with Confidence | 5th Edition⏎AUTHOR: Gray Morris, Deborah C.⏎PUBLISHER: Else

  4. #4 peak

    th Edition⏎This is Herb's step-by-step introduction to Java, updated for⏎Java

  5. #5 peak

    Chegg Com⏎Derivation Of Kir<|endoftext|>The poisonous Gartraps are the bane

  6. #6 peak

    appropriate action is taken to protect the health of all Albertans.⏎You can protect yourself from all coronaviruses by:⏎- Cleaning your hands frequently using either soap and warm water or alcohol-based hand sanitizers,⏎- Covering your cough and

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: if that, and if the following is true, then I can use the following and then I can have a certain degree of certainty, or 1, and 1, etc, and if that, and I can have the top,

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

    lands and produce but can also punish the whole community for the act of an individual.⏎Into the bargain, the government has gone a step further and shifted the onus of proving innocence on the accused; the person is presumed to be guilty until they

  2. #2 peak

    least restrictive way for the government to further that interest.⏎Minor costs or inconveniences imposed on religious institutions are insufficient to trigger RLUIPA’s protections. The burden must be “substantial.” And, likewise, once the institution has shown a

  3. #3 peak

    2014|⏎|ISBN: ||9781312744967 (DRM-EPUB)|⏎

  4. #4 peak

    land to grow crops and raise livestock–“allowing less to be retained or restored as natural habitat” the<|endoftext|>Hardcover ISBN: 978150360048

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. L3:1904 weakly excites
  2. L0:2015 weakly suppresses
  3. L5:46 weakly excites
  4. L3:2090 weakly excites
  5. L6:1291 weakly excites
  6. L5:28 weakly excites
  7. L1:2836 weakly excites
  8. L6:1476 weakly excites
  9. L6:1797 weakly suppresses
  10. L6:2885 weakly suppresses
  11. L5:180 weakly suppresses
  12. L6:2570 weakly suppresses
  13. L5:1495 weakly excites
  14. L5:94 weakly suppresses
  15. L3:1065 weakly excites
  16. L4:1745 weakly excites
  17. L6:106 weakly excites
  18. L1:2922 weakly excites
  19. L0:1333 weakly suppresses
  20. L6:1792 weakly suppresses
  21. L2:70 weakly excites
  22. L3:1545 weakly suppresses
  23. L3:2426 weakly excites
  24. L2:1334 weakly suppresses

Feeds later layers

  1. L15:1815 suppresses
  2. L8:1765 excites
  3. L10:1775 excites
  4. L9:2475 weakly excites
  5. L25:1250 weakly excites
  6. L22:2353 weakly excites the digit 0 in years 2000 to 2009
  7. L20:535 weakly suppresses
  8. L9:42 weakly excites
  9. L13:2463 weakly suppresses
  10. L8:1386 weakly excites
  11. L9:156 weakly suppresses
  12. L8:2995 weakly suppresses
  13. L8:1749 weakly suppresses
  14. L24:1594 weakly suppresses
  15. L8:1100 weakly excites
  16. L16:2752 weakly suppresses
  17. L11:2539 weakly suppresses
  18. L25:558 weakly suppresses
  19. L25:512 weakly suppresses
  20. L9:2141 weakly suppresses
  21. L27:2219 weakly suppresses
  22. L10:769 weakly suppresses
  23. L22:2182 weakly suppresses
  24. L11:1860 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 -1.29 and -1.221 values between -0.69 and -0.618 values between -0.61 and -0.5417 values between -0.54 and -0.4650 values between -0.46 and -0.3890 values between -0.38 and -0.31197 values between -0.31 and -0.23354 values between -0.23 and -0.16586 values between -0.16 and -0.08868 values between -0.08 and -0.01862 values between -0.01 and 0.07475 values between 0.07 and 0.15252 values between 0.15 and 0.2291 values between 0.22 and 0.3037 values between 0.30 and 0.3714 values between 0.37 and 0.452 values between 0.45 and 0.522 values between 0.52 and 0.601 values between 1.66 and 1.74offnegativepositiveoften

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. 鲞
  2. ureka
  3. ứt
  4. ette
  5. 这些问题
  6. udas
  7. ruptcy
  8. busters
  9. 钯
  10. '](

Suppresses

  1. HomeController
  2. MainFrame
  3. enus
  4. fore
  5. sov
  6. 続けて
  7. -controls
  8. ctype
  9. ствовать
  10. conv