Neuron Explorer

Qwen3-0.6B

'of' after abstract nouns

A neuron in layer 5 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 the word 'of' when it follows an abstract noun like 'viability', 'role' or 'popularity', as in 'the role of'. It switches off, and even goes negative, on 'of' in names like 'Party of Michigan' or body parts like 'muscle of the heart'. Turning it up made the model's writing drift into long, wordy questions and problem statements, which doesn't clearly relate to the trigger.

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

    to ensure the viability of the element;⏎- R.4: The nomination describes wide participation of the communities

  2. #2 peak

    internal P load is a direct method, making it a valuable tool for lake management and restoration.⏎Laboratory incubations of sediment cores can help determine the relative importance of internal vs. external P loads, as well as be used to answer a variety

  3. #3 peak

    against a loss of bee diversity and abundance. Our results will contribute to understanding the mechanisms that determine the role of native biodiversity on crop pollination and provide key information to farmers and conservation managers in

  4. #4 peak

    several years with an increase in the popularity of aromatherapy. Natural essential oils stimulate our mind and body, promote circulation and

  5. #5 peak

    flight was made at Trout Lake, Wisconsin. Aviation pioneer and wealthy Chicago sportsman Logan “Jack” Vilas made the initial flight in order to demonstrate the viability of using aircraft in fire prevention. Wisconsin’s Chief Forester Edward Griffith hired him but

  6. #6 peak

    keeping missions, as the result of military service, or on humanitarian missions.⏎ACT Planning and Land Authority: The ACT Planning and Land Authority is responsible for the naming of public places in the ACT (nomenclature). Public places include districts, divisions

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: I have been told by my mother, I have been told by my mother, I have been told by my mother, I have been told by my mother, I have been told by my mother, I have been told by my mother,

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

    of the Socialist Party of Michigan were, in succession, cigarmaker Ben blumberg of Grand Rapids

  2. #2 peak

    , the outgoing National Executive Committee of the Socialist Party of America expelled the militantly Marxist Socialist Party of Michigan with its approximately 6.

  3. #3 peak

    the lower number indicating diastolic blood pressure.⏎• Systolic blood pressure represents the pressure exerted by the blood on the arteries when the muscle of the heart contracts and pumps blood through the blood vessels to the rest of the body

  4. #4 peak

    bronze statue embodies the “development of Marxism” and shows respect to the revolutionary philosopher.⏎“Marx has significantly promoted the development of human society. The Communist Party of China has adopted Marx’s ideas, developing Marxism and using it in real life ...

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. L4:315 suppresses
  2. L4:1686 excites
  3. L4:4 suppresses
  4. L4:2711 suppresses
  5. L4:362 suppresses
  6. L3:537 weakly excites
  7. L0:1220 weakly suppresses
  8. L1:725 weakly suppresses "to" in "referred to"
  9. L3:1285 weakly suppresses
  10. L4:2347 weakly suppresses
  11. L3:315 weakly excites 'of' after a noun phrase
  12. L4:1151 weakly suppresses
  13. L1:632 weakly suppresses
  14. L2:895 weakly suppresses
  15. L3:1379 weakly excites
  16. L4:430 weakly excites
  17. L1:1041 weakly suppresses
  18. L0:2721 weakly excites
  19. L4:2801 weakly excites
  20. L1:3018 weakly excites
  21. L2:1321 weakly suppresses
  22. L2:927 weakly excites
  23. L3:74 weakly excites
  24. L2:2718 weakly suppresses

Feeds later layers

  1. L6:285 suppresses
  2. L15:477 weakly excites
  3. L18:1745 weakly suppresses
  4. L8:685 weakly suppresses
  5. L6:260 weakly suppresses
  6. L27:2356 weakly suppresses
  7. L9:1909 weakly excites
  8. L24:419 weakly excites
  9. L19:714 weakly suppresses
  10. L12:2853 weakly suppresses
  11. L12:459 weakly excites
  12. L10:1839 weakly suppresses
  13. L16:284 weakly excites
  14. L9:1299 weakly suppresses
  15. L17:2192 weakly suppresses
  16. L22:1239 weakly suppresses
  17. L13:1108 weakly suppresses
  18. L21:2547 weakly excites
  19. L6:153 weakly suppresses
  20. L13:2563 weakly suppresses
  21. L20:2805 weakly excites
  22. L19:713 weakly excites
  23. L14:534 weakly suppresses
  24. L16:2553 weakly 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.

2 values between -3.45 and -3.252 values between -3.04 and -2.841 values between -2.84 and -2.631 values between -2.63 and -2.421 values between -2.22 and -2.011 values between -2.01 and -1.813 values between -1.81 and -1.603 values between -1.60 and -1.391 values between -1.39 and -1.195 values between -1.19 and -0.9810 values between -0.98 and -0.779 values between -0.77 and -0.5718 values between -0.57 and -0.36243 values between -0.36 and -0.162328 values between -0.16 and 0.051146 values between 0.05 and 0.2681 values between 0.26 and 0.4617 values between 0.46 and 0.679 values between 0.67 and 0.876 values between 0.87 and 1.081 values between 1.08 and 1.294 values between 1.29 and 1.493 values between 1.49 and 1.702 values between 1.70 and 1.902 values between 1.90 and 2.112 values between 2.11 and 2.321 values between 2.32 and 2.521 values between 2.52 and 2.731 values between 2.73 and 2.941 values between 2.94 and 3.141 values between 3.35 and 3.551 values between 3.76 and 3.971 values between 4.58 and 4.79offnegativepositiveoften

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. 被执行
  2. coins
  3. 晶体
  4. coins
  5. ativos
  6. ünd
  7. rais
  8. $:
  9. CHED
  10. isci

Suppresses

  1. ersen
  2. ince
  3. uracy
  4. omore
  5. 让消费者
  6. erno
  7. kest
  8. er
  9. 人身
  10. quals