Neuron Explorer

Qwen3-0.6B

'would' in hypothetical or past statements

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 word 'would' when it talks about what something would have done or been, as in 'where the boiler would stand'. When it was turned down, the model got stuck repeating 'would', which fits the same word.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    120 GW (gigawatts) worth of coal power — the expected capacity of the cancelled plants had they been built and completed — an amount that would have been excessive anyway.⏎According to China’s 5-year plan, their target

  2. #2 peak

    Twinkle, twinkle little star….” Actually, however, some of them (like the Sun) are so big that one million planets the size of Earth would fi t inside them. Other stars are large enough to hold several stars the size

  3. #3 peak

    word "dressed" in a Welsh hat, a shorter version of the stovepipe, and a full-face piece of lacing, tied under where the chin would be.⏎The "descender" of the G was wrapped in a shawl

  4. #4 peak

    , it is far from over. The wreckage site is still being explored and the most valuable part of the ship (the stern castle, where the captain’s cabin would have concealed the most precious cargos of all) has yet to be located.⏎

  5. #5 peak

    wear a long piece of wood or metal in an attempt to straighten their back.⏎Dr Lund said: "Scoliosis is a painful illness, and Richard would have been in quite a lot of pain on a daily basis.⏎"These methods could

  6. #6 peak

    The central heat-recovery heat exchanger that makes the once-through air system efficient is the largest piece in the equipment room, taking pride of place where the boiler would traditionally stand.⏎Additional heating is provided by a cost-effective wood-pellet heater

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: In a wouldasammedarsatingly and aprisingly, the wouldasammedarsatingly wouldasitatingly…...Hmmly新时代新时代新时代新时代...新时代到了新时代到了到了到了到了到了到了到了到了到了到了到了到了到了

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

    -ready anytime soon. But, as Rashid writes, it could be the start of something:⏎The results are still not perfect, and there is still much work to be done, but the technology is very promising, and we hope that in a

  2. #2 peak

    5.For decades there has been talk about ‘urban transportation system⏎6.Raifroads have been largely replaced by automobiles and aviation.⏎Мы будем вынуждены удалить ваши ком

  3. #3 peak

    to overcome a myriad of hurdles. plants use it to sprout branches no one gets covered and each gets maximum amount of sunlight, and as<|endoftext|>Помогите пожаааалуйста, сам не справляюсьY.

  4. #4 peak

    to coast to celebrate the progress we’ve made in advancing gender equality in Canada, while reflecting on the work that remains to make sure that everyone – regardless of their gender – can reach their full potential.⏎

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. L21:853 strongly excites 'can' after 'and' or a noun subject
  2. L20:636 strongly suppresses
  3. L21:1657 strongly suppresses
  4. L22:627 strongly suppresses
  5. L23:505 strongly suppresses
  6. L23:373 strongly suppresses 'to' after a noun or adjective
  7. L23:257 strongly suppresses
  8. L21:1521 strongly excites
  9. L21:958 strongly excites
  10. L22:479 suppresses
  11. L23:30 suppresses
  12. L18:736 excites
  13. L19:1678 excites
  14. L17:1297 excites
  15. L19:858 suppresses
  16. L19:3044 suppresses
  17. L21:1708 excites
  18. L19:545 excites
  19. L19:1208 suppresses
  20. L22:2258 suppresses
  21. L21:1441 excites
  22. L23:697 suppresses
  23. L20:2089 suppresses
  24. L21:600 suppresses

Feeds later layers

  1. L25:364 strongly excites
  2. L26:296 strongly excites
  3. L26:225 strongly excites 'and' joining two abstract nouns
  4. L27:247 strongly excites
  5. L27:201 strongly excites
  6. L27:401 strongly excites
  7. L27:515 strongly excites
  8. L27:880 strongly excites
  9. L26:675 strongly excites
  10. L27:900 strongly excites
  11. L26:671 strongly excites
  12. L27:559 strongly excites
  13. L27:963 strongly excites
  14. L26:597 strongly excites
  15. L26:769 excites
  16. L26:862 excites
  17. L27:488 excites
  18. L25:181 excites
  19. L27:520 excites
  20. L26:623 excites
  21. L25:600 excites
  22. L27:801 excites
  23. L27:645 excites
  24. L26:50 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.

1907 values between -3.81 and -0.831757 values between -0.83 and 2.1667 values between 2.16 and 5.1528 values between 5.15 and 8.1317 values between 8.13 and 11.1210 values between 11.12 and 14.103 values between 14.10 and 17.095 values between 17.09 and 20.081 values between 20.08 and 23.064 values between 23.06 and 26.059 values between 26.05 and 29.037 values between 29.03 and 32.026 values between 32.02 and 35.0110 values between 35.01 and 37.9915 values between 37.99 and 40.9812 values between 40.98 and 43.967 values between 43.96 and 46.955 values between 46.95 and 49.936 values between 49.93 and 52.922 values between 52.92 and 55.914 values between 55.91 and 58.891 values between 58.89 and 61.886 values between 61.88 and 64.863 values between 64.86 and 67.854 values between 67.85 and 70.843 values between 70.84 and 73.821 values between 73.82 and 76.811 values between 79.79 and 82.782 values between 82.78 and 85.772 values between 91.74 and 94.721 values between 94.72 and 97.711 values between 97.71 and 100.701 values between 112.64 and 115.63offnegativepositiveoften

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. have
  2. involve
  3. vary
  4. satisfy
  5. occur
  6. operate
  7. require
  8. remind
  9. relate
  10. resemble

Suppresses

  1. 来了
  2. 给了我
  3. 打开了
  4. 给了
  5. 发现了
  6. 开了
  7. 选择了
  8. Didn
  9. 多了
  10. 会会长