Neuron Explorer

Qwen3-0.6B

'in' after a noun phrase

A neuron in layer 24 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 responds to the small word 'in' (and sometimes 'on') when it follows a noun or phrase and starts a qualifier, like 'discriminatory either of itself or in its effect' or 'selected in order to'. When it was turned up, the model repeated words like 'light view' and 'reason', though that link is loose.

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

    , creed, sex or marital status ⏎Further Article 23(1) states that “no law shall make any provision that is discriminatory either of itself or in its effect”.⏎The constitutional anti-discrimination clause is set out in Article

  2. #2 peak

    in price.⏎A measure of a variable's sensitivity to a change in another variable is elasticity. In economics, elasticity refers the degree to which individuals change their demand in response to price or income changes.⏎<|endoftext|>The liver is an essential organ of our

  3. #3 peak

    law could be the “Responsibility to Protect.” This principle, originally evolved against a background of genocide, tries to make it incumbent on states to take protective measures in events where human rights are imperilled.

  4. #4 peak

    This is even more true, under the current circumstances with Covid-19 and the current restrictions on people's movements and the heavy blow to SIDS’ economies in their continued reliance on tourism."⏎Vanessa Gray, Head of the Division for Least

  5. #5 peak

    , number, power, location, etc…), the installation may include additional substations designed "Secondary substations". The locations of these substations are carefully selected in order to optimize the budget dedicated to MV and LV power cables. They are supplied

  6. #6 peak

    about the application of the law of armed conflict to hostilities in the space domain. Efforts to achieve greater clarity must be undertaken before such hostilities occur, in part because such efforts will help to address some of the current challenges in space security

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: for in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in

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

    course, combining these free tools to deliver learning over the web is not without its challenges. One significant challenge is that of the level

  2. #2 peak

    - Excellent cutting surface roughness can be produced even when machining at low speed.⏎- Suitable for soft materials or workpieces<|endoftext|>Fernanda vertelt haar verhaal in een gesprek met CSL Behring⏎Living With Alpha 1

  3. #3 peak

    you don’t do well, it’s not just your failure it’s the extended family as well.⏎Marrying an unstable husband and having four of her five in five years would stress anyone, but Flora still wanted her children to have the

  4. #4 peak

    a node is visited __before__ its children. In "post-order" traversal, a node is visited __after__ its children. The more rarely used "in-order" traversal is generally applicable only to binary trees, and is where you visit

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. L20:2850 excites
  2. L17:706 excites
  3. L17:303 suppresses
  4. L19:704 excites
  5. L18:293 suppresses
  6. L20:283 excites
  7. L12:234 excites
  8. L5:763 suppresses
  9. L16:1059 weakly excites
  10. L16:1557 weakly excites 'in' meaning 'in this case' or 'in between'
  11. L4:289 weakly excites
  12. L14:1625 weakly excites the word 'in' after a noun
  13. L10:132 weakly excites
  14. L22:535 weakly excites
  15. L15:771 weakly suppresses
  16. L19:2866 weakly suppresses
  17. L17:1721 weakly excites 'by' introducing a method or means
  18. L19:1464 weakly suppresses
  19. L23:1366 weakly excites
  20. L23:1506 weakly excites
  21. L15:906 weakly suppresses
  22. L15:675 weakly excites
  23. L9:1358 weakly excites
  24. L3:232 weakly excites 'on' after a noun like 'impact on'

Feeds later layers

  1. L27:1684 excites
  2. L27:814 excites
  3. L27:1282 excites
  4. L27:1809 excites
  5. L27:478 weakly excites
  6. L27:2423 weakly excites
  7. L26:346 weakly excites
  8. L27:1355 weakly suppresses
  9. L27:2053 weakly suppresses
  10. L26:657 weakly suppresses the word 'people' coming next
  11. L26:110 weakly suppresses
  12. L26:1764 weakly suppresses
  13. L25:1085 weakly excites
  14. L27:1846 weakly excites
  15. L27:1199 weakly excites
  16. L25:1526 weakly excites
  17. L25:390 weakly excites
  18. L26:1530 weakly excites
  19. L27:208 weakly excites
  20. L27:1810 weakly suppresses
  21. L27:1424 weakly suppresses
  22. L27:3004 weakly excites
  23. L27:785 weakly excites
  24. L25:50 weakly suppresses 'di' syllable in diabetes and diet words

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 -14.74 and -13.243 values between -13.24 and -11.745 values between -8.74 and -7.243 values between -7.24 and -5.746 values between -5.74 and -4.2422 values between -4.24 and -2.7472 values between -2.74 and -1.243284 values between -1.24 and 0.26383 values between 0.26 and 1.7643 values between 1.76 and 3.2725 values between 3.27 and 4.7712 values between 4.77 and 6.2713 values between 6.27 and 7.7710 values between 7.77 and 9.273 values between 9.27 and 10.775 values between 10.77 and 12.272 values between 12.27 and 13.773 values between 13.77 and 15.271 values between 15.27 and 16.771 values between 16.77 and 18.273 values between 18.27 and 19.771 values between 19.77 and 21.271 values between 21.27 and 22.771 values between 22.77 and 24.271 values between 24.27 and 25.771 values between 25.77 and 27.272 values between 27.27 and 28.781 values between 43.78 and 45.28offnegativepositiveoften

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. light
  2. view
  3. Light
  4. light
  5. (light
  6. relation
  7. connection
  8. virtue
  9. →light
  10. Light

Suppresses

  1. esimal
  2. stile
  3. ego
  4. olang
  5. ducible
  6. oodles
  7. TOCOL
  8. es
  9. 禀
  10. <Any