Neuron Explorer

Qwen3-0.6B

'War' after 'World'

A neuron in layer 12 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 'War' in phrases like 'World War I' and 'World War II', where it is the word right after 'World'. Turning it up did not make the model talk about wars, so the text it produces does not show this link.

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

    exciting phases in European contemporary history, including its fine art development. It comprises the last quarter of the 19th century and the decades leading up to World War II. The era is characterized by dramatic transformations in not just technological and industrial developments

  2. #2 peak

    another ship from another war would join the Cormoran in its water Apra Harbor grave. The Tokai Maru was a Japanese passenger-cargo ship before World War II, but after the outbreak of World War II it became a transport for the

  3. #3 peak

    Consequently, the South became a place of economic and racial turmoil, and African Americans were desperate to find some relief. The promise of jobs in northern factories during World War I provided African Americans an opportunity to escape the harsh realities of the South. Between

  4. #4 peak

    in 1824, and Singapore became one of the British Straits Settlements in 1826. Occupied by the Japanese during World War II, Singapore became independent from the United Kingdom in 1963 and

  5. #5 peak

    yarrow. It’s used as a curative remedy for malaria, jaundice, and liver disorders throughout the world.⏎Yarrow was even used during World War II in England when the country experienced a shortage of needed medical supplies. Officials encouraged

  6. #6 peak

    leagues' first night baseball game was played in Cincinnati. (The Reds defeated the Philadelphia Phillies 2-1.)⏎In 1941, in World War II, the German battleship Bismarck sank the British battle cruiser Hood

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: the number of a person and the number of a person and the number of their age (in years) and their gender (in their gender) and their sex (in their gender) and their gender (in their gender) and their sex

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

    was inaccurate, or both. Most efforts focused on option two since there were several other observations of the Sun that supported the current solar models. To explain the solar neutrino problem, one theory is that neutrinos undergo oscillations. Simply put,

  2. #2 peak

    Yes, you heard it right!⏎How Low-Carb Diet Risks Heart-Health:⏎A study, presented at the European Society of Cardiology (ECC) Congress 2018, found low-car

  3. #3 peak

    a given line not through the focus, called the directrix, a parabola is the locus of points such that the distance to the focus equals the distance to the directrix. [Adopt

  4. #4 peak

    is considered to be an approximation of the other. The MSE can be described as the mean of the square of the differences in the pixel values between the corresponding pixels of the two images.⏎Express

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. L6:1693 weakly suppresses
  2. L7:2766 weakly excites
  3. L4:543 weakly excites
  4. L10:1281 weakly suppresses
  5. L7:743 weakly excites
  6. L6:2033 weakly excites
  7. L7:555 weakly suppresses
  8. L8:1427 weakly suppresses
  9. L4:2442 weakly suppresses
  10. L7:2104 weakly suppresses
  11. L9:2863 weakly suppresses
  12. L5:1396 weakly excites
  13. L7:2727 weakly suppresses
  14. L6:2499 weakly suppresses
  15. L5:2244 weakly excites
  16. L7:2715 weakly excites
  17. L2:175 weakly excites
  18. L11:1711 weakly suppresses
  19. L10:1670 weakly excites
  20. L9:2095 weakly excites
  21. L9:2932 weakly excites
  22. L6:2665 weakly suppresses
  23. L10:1833 weakly excites
  24. L3:129 weakly excites

Feeds later layers

  1. L17:2227 weakly excites
  2. L13:912 weakly suppresses
  3. L27:2738 weakly excites unclear: fragments of split words
  4. L15:1425 weakly suppresses
  5. L24:2174 weakly excites
  6. L25:2213 weakly excites
  7. L15:2727 weakly suppresses
  8. L13:172 weakly excites
  9. L16:2367 weakly excites
  10. L14:1002 weakly excites
  11. L19:2853 weakly excites
  12. L13:2977 weakly suppresses
  13. L22:1878 weakly excites
  14. L16:2820 weakly excites
  15. L14:1875 weakly suppresses
  16. L25:801 weakly suppresses
  17. L24:205 weakly suppresses
  18. L20:1416 weakly excites
  19. L21:682 weakly excites
  20. L14:843 weakly suppresses
  21. L13:1584 weakly suppresses
  22. L19:599 weakly suppresses
  23. L13:393 weakly suppresses
  24. L20:1048 weakly suppresses word before a quantity or number

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 -2.21 and -2.111 values between -0.95 and -0.851 values between -0.64 and -0.537 values between -0.53 and -0.4319 values between -0.43 and -0.32100 values between -0.32 and -0.22376 values between -0.22 and -0.111162 values between -0.11 and -0.011570 values between -0.01 and 0.10474 values between 0.10 and 0.20143 values between 0.20 and 0.3122 values between 0.31 and 0.4113 values between 0.41 and 0.523 values between 0.52 and 0.6210 values between 0.62 and 0.733 values between 0.73 and 0.831 values between 0.94 and 1.041 values between 1.46 and 1.571 values between 1.88 and 1.98offnegativepositiveoften

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. 为企业
  2. UNITY
  3. itty
  4. Grimm
  5. .Microsoft
  6. iverse
  7. ADX
  8. verity
  9. haust
  10. Omar

Suppresses

  1. done
  2. nameLabel
  3. one
  4. ones
  5. onas
  6. asts
  7. ste
  8. Capitals
  9. SetName
  10. ock