Neuron Explorer

Qwen3-0.6B

"of" in "of use/concern/interest" phrases

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 lights up on the word "of" in set phrases like "of interest", "of use" or "of concern", right after words like "be" or "are". When it was turned up, the model got stuck repeating words like "know", and when turned down it produced odd filler words like "reason" and "confidence".

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

    sort of computer that take advantage of aspects of quantum physics to solve certain sorts of problems dramatically faster than conventional computers can. While they will, no doubt, be of huge

  2. #2 peak

    for individual and scientific applications. Often times, we need to use additional converters because different metric units are used. In this case, kilometers into meters calculator may be of use to you.⏎Converting long rivers:⏎- Nelson–Saskatchewan

  3. #3 peak

    earth.⏎We have looked at carbon dioxide separately since it has been the major headline-producing greenhouse gas.⏎But there are a number of other green house gases that are of concern. Weight for weight some of these have a much more potent effect than carbon

  4. #4 peak

    of time in which to contemplate one’s own unhappiness. The effect of the conformity demanded by society on the psychological health or spiritual wellbeing of its constituents is also of interest, especially to some of the oldest theorists in this school, the Transc

  5. #5 peak

    A few studies of human populations have suggested that fluoride might be associated with alterations in reproductive hormones, fertility, and Down's syndrome, but their design limitations make them of little value for risk evaluation⏎- Assessing whether fluoride constitutes a risk factor for

  6. #6 peak

    that we are likely to have in the coming decade, a debate about how we know the past and how we look at the written record that I suspect will be of interest to literary scholars and historians alike. It is a philosophical debate about idealism

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: power power power. power power power. power power power. power power power. power power power. power power power. power power power. power power power. power power power. power power power. power power power. power power power.

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

    and Balances⏎The process of creating the Constitution was a real challenge. Everyone involved had concerns about how the government would be run and what person or group would have power. One person who heavily influenced the idea of checks and balances was the Baron

  2. #2 peak

    especially incandescent and vfixtures can be a fire hazard.⏎Permits and Inspections⏎Get a wiring permit. The National Electric Code and the local Authority Having Jurisdiction permits the homeowner to do everything that a licensed electrician can do,

  3. #3 peak

    -related issues (tardiness). In the hospitality industry, it can be poor customer service, safety violations, etc.⏎As per the due process, an employee has a right to be informed of his/her

  4. #4 peak

    of some Providential call upon⏎us. We ought never to take for granted such a great event as the uprising⏎of the British Empire. But we have no right to this position of special⏎and honorable responsibility unless we are determined to use

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. L19:1727 excites
  2. L16:1860 suppresses
  3. L21:1060 suppresses
  4. L19:2406 excites
  5. L22:2587 suppresses
  6. L17:2009 suppresses
  7. L22:3060 excites
  8. L22:554 suppresses
  9. L20:1596 excites
  10. L21:1813 excites
  11. L17:2436 suppresses
  12. L23:203 suppresses
  13. L22:809 excites
  14. L4:309 excites
  15. L17:2363 weakly suppresses
  16. L16:1619 weakly suppresses
  17. L18:2815 weakly suppresses
  18. L21:2448 weakly suppresses "a" after give or get verbs
  19. L18:2711 weakly suppresses
  20. L18:3063 weakly excites
  21. L20:2525 weakly suppresses
  22. L19:3005 weakly excites
  23. L4:291 weakly suppresses
  24. L3:2992 weakly excites

Feeds later layers

  1. L26:298 suppresses
  2. L25:99 suppresses
  3. L26:944 suppresses
  4. L26:528 weakly suppresses
  5. L26:501 weakly suppresses
  6. L26:1731 weakly suppresses
  7. L25:1190 weakly suppresses
  8. L27:249 weakly excites
  9. L26:1318 weakly suppresses
  10. L27:1296 weakly suppresses
  11. L26:2228 weakly suppresses
  12. L26:191 weakly suppresses
  13. L25:1908 weakly suppresses
  14. L27:623 weakly suppresses
  15. L25:284 weakly excites
  16. L26:1703 weakly suppresses
  17. L25:125 weakly excites
  18. L26:2726 weakly suppresses
  19. L25:1133 weakly suppresses
  20. L26:1088 weakly excites
  21. L27:735 weakly suppresses
  22. L26:2617 weakly suppresses
  23. L26:1149 weakly suppresses
  24. L27:1855 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 -32.63 and -31.541 values between -30.44 and -29.351 values between -21.72 and -20.631 values between -16.27 and -15.181 values between -15.18 and -14.091 values between -14.09 and -13.001 values between -11.91 and -10.822 values between -10.82 and -9.736 values between -9.73 and -8.648 values between -8.64 and -7.558 values between -7.55 and -6.4610 values between -6.46 and -5.368 values between -5.36 and -4.2716 values between -4.27 and -3.1827 values between -3.18 and -2.0983 values between -2.09 and -1.001628 values between -1.00 and 0.091962 values between 0.09 and 1.1875 values between 1.18 and 2.2723 values between 2.27 and 3.3614 values between 3.36 and 4.4511 values between 4.45 and 5.546 values between 5.54 and 6.635 values between 6.63 and 7.722 values between 7.72 and 8.813 values between 8.81 and 9.904 values between 9.90 and 10.99offnegativepositiveoften

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. ssp
  2. knew
  3. 疾
  4. Known
  5. вин
  6. emen
  7. ôm
  8. vik
  9. äter
  10. 周刊

Suppresses

  1. 權利
  2. reason
  3. confidence
  4. 权力
  5. 權
  6. recht
  7. 权利
  8. 权限
  9. cause
  10. 理由