Neuron Explorer

Qwen3-0.6B

"and" or "or" linking two counted items

A neuron in layer 14 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 "and" or "or" when it joins two things in a pair or list, especially when a number of people or things is being described, like "20 percent of children and teens". Turning it up did not produce clearly related text.

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

    do more to protect students from online attacks.⏎There's only so much lawmakers can do to curb the problem, however.⏎Officials estimate 20 percent of children and teens are victims of this 21st century form of school-yard teasing.⏎

  2. #2 peak

    atosphere.”⏎The length of the iceberg from one end to the next is 99.4 miles which is as much as 58 Golden Gate Bridges joined end to end.⏎The lead scientist from Swansea University, UK, investigating the Lars

  3. #3 peak

    made up the berg was between 3,000 and 6,000 years old.⏎Of the approximately 2,200 passengers and crew aboard the Titanic, less than half survived. There were only 20

  4. #4 peak

    effective way to burn more calories, which helps you lose weight. If calories consumed in food are less than calories spent during exercise and other daily activities, you will lose weight. It is impossible not to! Many

  5. #5 peak

    epidemic by comparing flu infections to those for the same week in previous years. Last week, 9.8 percent of all people who died succumbed to pneumonia and influenza – above the epidemic threshold of 7.3 percent for that week.⏎

  6. #6 peak

    weather maps to distinguish the items with the most significant increases or decreases in inflation from those with less notable changes.⏎The map’s horizontal axis displays the twenty-three goods and services into which we have divided the overall CPI basket. In green are the food

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: which is about a mathematical concept of a number, which is a mathematical concept of a real number, which is a mathematical concept of a number, which is a real number, which is a property of a number, about a property of a

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 how many dimensions there really are, emotions are typically more simple. We tend to distinguish between two dimensions of emotions: arousal and direction (or valence, or motivation). Arousal responses signify the intensity of emotions.

  2. #2 peak

    . An endodontist provides treatment for:⏎- Chipped teeth⏎- Broken teeth⏎- Infected teeth⏎- Cracked teeth⏎- Inflamed/infected dental pulp⏎- Dental abscess⏎These are only some of the

  3. #3 peak

    exist two different realities for every individual that lives; that which is physical, referred to as the body, and that which is abstract, referred to as the soul or the mind. According to Plato,

  4. #4 peak

    get reactivated any time. When the HSV-1 is reactivated, cold sores appear. Some of the more common triggers for its reactivation and, hence, the formation of cold sores, include the following.⏎- Psychological and emotional

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. L9:2010 suppresses
  2. L9:738 suppresses
  3. L8:70 excites 'and' between stacked superlatives
  4. L13:1561 excites
  5. L10:144 suppresses
  6. L13:637 suppresses opening parenthesis after a term
  7. L8:933 weakly suppresses words introducing an example or aside
  8. L6:912 weakly suppresses
  9. L11:2774 weakly excites 'least' in 'at least' hedges
  10. L12:1283 weakly excites
  11. L10:1643 weakly suppresses
  12. L10:1392 weakly suppresses
  13. L6:1373 weakly excites
  14. L12:2930 weakly suppresses "and" joining contrasting pairs
  15. L2:1055 weakly excites
  16. L0:611 weakly suppresses
  17. L12:1504 weakly suppresses
  18. L11:2212 weakly suppresses
  19. L7:428 weakly excites
  20. L6:427 weakly suppresses
  21. L1:342 weakly suppresses
  22. L6:1097 weakly suppresses
  23. L4:486 weakly excites
  24. L12:1142 weakly suppresses

Feeds later layers

  1. L16:2632 suppresses
  2. L15:1760 suppresses
  3. L15:2033 suppresses
  4. L16:1065 suppresses
  5. L15:1501 suppresses "and" joining two parts of a name or phrase
  6. L18:1386 suppresses unclear: words before 'and' or 'or'
  7. L17:727 weakly suppresses
  8. L16:1856 weakly excites
  9. L15:2205 weakly excites
  10. L16:1074 weakly suppresses
  11. L17:2274 weakly suppresses
  12. L17:540 weakly suppresses
  13. L15:217 weakly excites
  14. L19:291 weakly suppresses 'and' joining two full clauses
  15. L25:123 weakly excites
  16. L17:2891 weakly excites
  17. L15:1236 weakly excites
  18. L19:1783 weakly suppresses
  19. L15:93 weakly excites
  20. L18:1083 weakly excites
  21. L16:1322 weakly suppresses
  22. L22:2779 weakly excites
  23. L26:2160 weakly excites
  24. L20:955 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.

1 values between -4.27 and -4.111 values between -3.95 and -3.802 values between -3.64 and -3.482 values between -3.32 and -3.171 values between -2.54 and -2.381 values between -2.06 and -1.901 values between -1.90 and -1.752 values between -1.43 and -1.273 values between -1.27 and -1.126 values between -1.12 and -0.967 values between -0.96 and -0.809 values between -0.80 and -0.6425 values between -0.64 and -0.4858 values between -0.48 and -0.33219 values between -0.33 and -0.171200 values between -0.17 and -0.011801 values between -0.01 and 0.15425 values between 0.15 and 0.3087 values between 0.30 and 0.4612 values between 0.46 and 0.6216 values between 0.62 and 0.788 values between 0.78 and 0.946 values between 0.94 and 1.097 values between 1.09 and 1.252 values between 1.25 and 1.412 values between 1.41 and 1.571 values between 1.72 and 1.883 values between 1.88 and 2.04offnegativepositiveoften

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. bumper
  2. trú
  3. ruise
  4. ราชการ
  5. adero
  6. Bermuda
  7. criptor
  8. ores
  9. 擦
  10. orge

Suppresses

  1. frames
  2. frames
  3. McG
  4. frame
  5. _frames
  6. eft
  7. Hob
  8. --)⏎
  9. roat
  10. --)