Neuron Explorer

Qwen3-0.6B

opening bracket before an acronym

A neuron in layer 23 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 an opening bracket that comes right after a full name, just before its short form, like "(NAMS)" or "(BAS)". Its strongest negative values fall on odd letters inside Finnish or other non-English words, so the pattern is mixed. Steering did not show a clear effect.

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

    last child, certain measurements may be linked with her projected lifespan, according to a study published Wednesday in Menopause, the journal of The North American Menopause Society (NAMS).⏎More specifically, leukocyte telomere length – telomeres

  2. #2 peak

    In early December, the East African Paediatric Association (EAPA) and Kenya Paediatric Assocation (KPA) in collaboration

  3. #3 peak

    another mother⏎The concept of digital twins, particularly when combined with MSPC, is relatively new, yet a well-constructed system based on Process Analytical Technology (PAT) will leverage both of these concepts, and as a result maximize your process

  4. #4 peak

    3⏎The British Safety Council International Safety Award 2011⏎Halley is the most southerly science research station operated by the British Antarctic Survey (BAS) and is located on the 150-metre thick floating

  5. #5 peak

    thread’s major diameter, minor diameter, pitch diameter, pitch (in case of metric threads) or threads per Inch (TPI in case of Standard American Equivalent (SAE) threads), flank angle, etc. Normally, a GO / NO

  6. #6 peak

    Policy, The Defence Advanced Research Projects Agency is developing some extremely high technology to tackle possible alien invasions, an example being the Extreme Accuracy Tasked Ordnance (EXACTO), a self-guided bullet, which aims to counter aliens moving

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 I don't depend on my first choice, I can choose my second choice. If I choose my second choice first, then I can choose my third choice, and so on, my choice depends on my first choice. I donap

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

    .⏎Metsäpohjainen piha paransi päiväkotilasten immuunijärjestelmää jo kuukaudessa<|endoftext|>Research Looks at Alternative Power for Military⏎Source

  2. #2 peak

    linkit|⏎|Tila||Julkaistu - 1 syysk. 2021|⏎|OKM-julkaisutyyppi||A1 Julkaistu artikkeli, soviteltu

  3. #3 peak

    adopted the Genocide Convention. It defined genocide as “acts committed with intent to destroy,<|endoftext|>1905 - 1938⏎György Nemes was a graphic artist, active during the 192

  4. #4 peak

    been working as a graphic designer. He designed posters, brochures, album covers, etc. He many times cooperated with other artists, for example György Rad

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. L4:2339 strongly excites
  2. L2:1699 strongly suppresses
  3. L18:249 strongly suppresses
  4. L5:779 strongly excites
  5. L19:390 strongly excites
  6. L19:287 excites
  7. L16:56 suppresses
  8. L2:1749 excites
  9. L18:221 excites
  10. L16:143 excites
  11. L0:49 suppresses
  12. L17:542 suppresses
  13. L1:20 excites
  14. L19:798 excites
  15. L1:141 suppresses
  16. L0:54 excites
  17. L17:886 excites
  18. L1:542 excites
  19. L6:401 suppresses
  20. L2:387 excites
  21. L10:2401 excites
  22. L18:924 weakly excites
  23. L1:56 weakly suppresses
  24. L16:17 weakly suppresses

Feeds later layers

  1. L27:1563 suppresses
  2. L24:2225 excites
  3. L24:725 excites
  4. L26:1203 weakly excites
  5. L26:789 weakly excites
  6. L25:1808 weakly excites
  7. L27:203 weakly suppresses
  8. L26:2111 weakly excites
  9. L27:2409 weakly excites
  10. L27:980 weakly excites
  11. L25:59 weakly excites
  12. L25:2710 weakly suppresses
  13. L24:156 weakly excites
  14. L27:1569 weakly excites first pieces of split words like clo and Le
  15. L27:2008 weakly excites
  16. L25:790 weakly suppresses
  17. L24:677 weakly excites
  18. L25:2358 weakly excites
  19. L24:90 weakly excites
  20. L24:94 weakly suppresses
  21. L25:194 weakly suppresses
  22. L27:1833 weakly excites
  23. L25:1564 weakly excites word pieces before 'trans' or 'transfer'
  24. L27:2136 weakly excites capital M inside abbreviations

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.

2 values between -4.69 and -4.094 values between -2.90 and -2.301 values between -2.30 and -1.716 values between -1.71 and -1.1151 values between -1.11 and -0.512565 values between -0.51 and 0.081205 values between 0.08 and 0.6840 values between 0.68 and 1.288 values between 1.28 and 1.877 values between 1.87 and 2.474 values between 2.47 and 3.075 values between 3.07 and 3.661 values between 3.66 and 4.262 values between 4.26 and 4.862 values between 4.86 and 5.451 values between 5.45 and 6.051 values between 6.05 and 6.651 values between 13.21 and 13.802 values between 18.58 and 19.17offnegativepositiveoften

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. 았다
  2. 았
  3. omat
  4. kah
  5. ól
  6. fang
  7. omas
  8. ัส
  9. ika
  10. mal

Suppresses

  1. ä
  2. ös
  3. ök
  4. äm
  5. öt
  6. öh
  7. än
  8. äll
  9. ön
  10. Latter