Neuron Explorer

Qwen3-0.6B

opening parenthesis after a noun

A neuron in layer 15 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 reacts to an opening bracket that comes right after a word, usually to add a short alternative, abbreviation or number, like 'child(ren)' or 'National Weather Service (NWS)'. Turning it up made the model's writing repetitive and loopy, with no 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

    of the Guardian ad Litem is to investigate the facts of the case for the Judge and report back to the Court – to talk with both parents, the child(ren) and other important witnesses in the case. The types of case where Guardian

  2. #2 peak

    the use of English and official language in the content classrooms, typically mathematics and science. According to Cummins, bilingual education is defined as “the used of two (or more) language of instruction at some point in a student’s school career”

  3. #3 peak

    ments: layers of accident-containing steel and concrete.⏎One of fission’s byproduct is heat. (Another byproduct may be gamma rays.) The heat boils (clean) water, producing steam, which turns a turbine and produces electricity. If

  4. #4 peak

    The real beauty of this method is<|endoftext|>In a divorce or other family law case, if the Judge finds that it is in the best interests of the minor child(ren) in the case, the Judge can appoint a “Guardian ad L

  5. #5 peak

    , heroes, and villains that ascribe meaning to history and purpose to current events. These are found location to location.⏎In his dissertation research, Matthews saw twelve (12) moderating variables emerge from metanarrative.

  6. #6 peak

    done to relieve the shortness of breath that a large pleural effusion can cause.⏎Normally the pleural cavity contains less than 20 milliliters (4 teaspoons) of clear, yellowish (

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: it's a new idea that I have not seen in a lot of people. It might be a very simple and simple to solve, but I don't know. I don that. I don't know. I'm not sure. 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

    change the look of Kentucky gardens for years to come, said Nicole Ward Gauthier, extension plant pathologist for the University of Kentucky College of Agriculture, Food and Environment.⏎Impatiens downy mildew was first reported in the state in

  2. #2 peak

    spirit and the letter of the law, recognise a legal problem and understand whether expert advice is required.⏎Who can advise me further?⏎The Department for Culture, Media and Sport has produced Guidance to help people understand the Dealing in Cultural Objects (

  3. #3 peak

    , the Ministry of Environment, Forest and Climate Change has come up with the first draft of the amendments to the Indian Forest

  4. #4 peak

    ian Hilfrich; Senior Lecturer in American History and Robert Mason, Professor of Twentieth-Century U.S. History from School of History, Classics and Archaeology. You can sign up for this free course here: The Making of

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. L14:2276 excites opening bracket after a term or number
  2. L4:493 excites
  3. L14:915 suppresses
  4. L14:730 suppresses
  5. L14:1073 excites
  6. L7:1310 excites
  7. L7:2976 suppresses
  8. L14:1924 weakly suppresses
  9. L13:2918 weakly suppresses
  10. L14:3 weakly excites
  11. L10:899 weakly excites
  12. L13:1942 weakly suppresses
  13. L14:500 weakly suppresses
  14. L14:2002 weakly suppresses
  15. L9:2547 weakly excites
  16. L10:1654 weakly excites opening parenthesis before an acronym
  17. L9:2019 weakly suppresses
  18. L11:558 weakly excites
  19. L13:1126 weakly excites
  20. L14:470 weakly excites
  21. L13:1770 weakly suppresses
  22. L6:924 weakly excites
  23. L11:1743 weakly excites parenthesis after prophet or Allah names
  24. L4:1142 weakly suppresses

Feeds later layers

  1. L16:665 suppresses
  2. L19:224 excites
  3. L16:480 suppresses
  4. L17:324 weakly excites
  5. L20:1380 weakly suppresses end of citations and titles
  6. L17:566 weakly excites
  7. L22:451 weakly suppresses 'to' before a verb of purpose
  8. L22:595 weakly suppresses the word 'number' after 'the'
  9. L24:2742 weakly suppresses
  10. L22:2530 weakly suppresses
  11. L17:582 weakly suppresses
  12. L22:1913 weakly suppresses
  13. L17:2093 weakly suppresses
  14. L16:450 weakly suppresses
  15. L24:293 weakly suppresses
  16. L16:337 weakly excites
  17. L16:1258 weakly excites
  18. L17:1612 weakly excites
  19. L16:627 weakly excites
  20. L19:1530 weakly excites
  21. L17:2286 weakly suppresses
  22. L16:743 weakly suppresses
  23. L24:336 weakly excites
  24. L16:2249 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.

1 values between -2.26 and -2.142 values between -1.80 and -1.692 values between -1.69 and -1.573 values between -1.46 and -1.356 values between -1.35 and -1.237 values between -1.23 and -1.126 values between -1.12 and -1.0024 values between -1.00 and -0.899 values between -0.89 and -0.7834 values between -0.78 and -0.6649 values between -0.66 and -0.5580 values between -0.55 and -0.43133 values between -0.43 and -0.32238 values between -0.32 and -0.21445 values between -0.21 and -0.091464 values between -0.09 and 0.02907 values between 0.02 and 0.14316 values between 0.14 and 0.2598 values between 0.25 and 0.3640 values between 0.36 and 0.489 values between 0.48 and 0.598 values between 0.59 and 0.714 values between 0.71 and 0.826 values between 0.82 and 0.934 values between 0.93 and 1.053 values between 1.05 and 1.161 values between 1.16 and 1.282 values between 1.28 and 1.391 values between 1.39 and 1.501 values between 1.50 and 1.623 values between 1.96 and 2.071 values between 2.07 and 2.191 values between 2.19 and 2.30offnegativepositiveoften

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. 太过
  2. ipsis
  3. 城镇
  4. خفض
  5. enko
  6. 帆
  7. -chan
  8. -san
  9. foon
  10. illin

Suppresses

  1. prise
  2. orners
  3. ass
  4. tempt
  5. berg
  6. atest
  7. 有个
  8. 率为
  9. ulated
  10. tổng