Neuron Explorer

Qwen3-0.6B

middle digits of numbers

A neuron in layer 15 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 a digit inside a number, most often a zero in a temperature, distance or speed such as 5800 or 280 nm. It is not a clear theme in the steered text, which drifted into questions and technical topics.

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

    than any of the planets or moons. The Sun's spectrum shows that it has a temperature of 5800 Celsius. The high temperature on the Sun's surface causes it to emit a great

  2. #2 peak

    Mars in several slingshot maneuvers to accelerate and help hurdle the spacecraft toward its destination. The comet is traveling toward the sun at 34,175 mph. After a high-speed chase across nearly 4 billion miles (6.

  3. #3 peak

    adopting the 20-20-20 rule: Take a 20-second break to look at something 20 feet away every 20 minutes.⏎This will help rest your eyes. Also, blink frequently to help keep

  4. #4 peak

    breweries, bottled water plants and pharmaceutical facilities.⏎UV-C light, often referred to as Germicidal UV, exists between 200 and 280nm and is therefore invisible to the naked eye. Naturally occurring UV-C rays from

  5. #5 peak

    at the asteroid 2005 YU55 a few days ago, when it passed the Earth at the relatively close distance of about 320,000 km. One of those eyes was actually in space as well

  6. #6 peak

    dawn and spring on Earth.⏎English is Germanic and therefore based on Proto-Indo-European (PIE) language. PIE was spoken around 3500 B.C.E. by a group of nomads called the Aryans living

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: what if I have a group of 10000 - 5000 - 5000 - 5000 - 5000 - 5000 - 500

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

    against the disease,” says UCSD’s Kim Barrett.⏎The team grew cancer cells in the laboratory and found that leptin could stimulate their growth. In two out of three cell lines, leptin also blocked normal programmed cell death, apoptosis, which

  2. #2 peak

    ypti. This 4-millimeter-long insect is responsible not only for the Zika outbreak but also for transmitting yellow fever and two other excruciating diseases, dengue and chikungunya, that

  3. #3 peak

    ven tongues like as of fire’. So, once again fire possess ambiguous symbolism: the element is both associated with both the Devil and hell, and God and the Holy<|endoftext|>Previous: [SVG]

  4. #4 peak

    keep working on several issues aimed at improving relations.⏎Trump and Kim held another face-to-face meeting last month in Vietnam. That meeting produced no major agreements. At both meetings, the leaders reportedly considered a proposal to open a liaison office in North Korea

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. L14:2008 excites
  2. L13:2099 suppresses
  3. L11:233 suppresses
  4. L14:826 suppresses
  5. L14:1601 suppresses
  6. L6:284 excites
  7. L13:1575 excites
  8. L10:167 suppresses
  9. L8:298 excites
  10. L14:2057 excites
  11. L13:330 suppresses
  12. L14:2291 suppresses
  13. L12:1065 excites
  14. L13:2887 suppresses
  15. L14:990 suppresses
  16. L11:1320 suppresses
  17. L14:2916 excites
  18. L13:1370 excites
  19. L13:2180 suppresses
  20. L13:1800 suppresses
  21. L13:1729 excites
  22. L13:2727 suppresses
  23. L13:2715 suppresses
  24. L14:2651 excites

Feeds later layers

  1. L16:2662 suppresses
  2. L16:199 excites
  3. L17:390 suppresses
  4. L16:474 suppresses
  5. L16:1327 suppresses
  6. L19:2647 weakly suppresses
  7. L16:297 weakly excites
  8. L16:1350 weakly suppresses
  9. L19:713 weakly excites
  10. L17:2057 weakly excites
  11. L16:1761 weakly suppresses
  12. L16:229 weakly excites
  13. L16:1633 weakly excites
  14. L19:2453 weakly suppresses
  15. L19:862 weakly excites
  16. L19:1496 weakly suppresses
  17. L19:249 weakly suppresses
  18. L19:2587 weakly suppresses
  19. L18:732 weakly excites
  20. L16:500 weakly excites
  21. L18:19 weakly excites
  22. L18:1143 weakly suppresses
  23. L17:565 weakly suppresses
  24. L16:1344 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 -3.40 and -3.121 values between -3.12 and -2.841 values between -2.57 and -2.293 values between -2.29 and -2.024 values between -2.02 and -1.7414 values between -1.74 and -1.4724 values between -1.47 and -1.1967 values between -1.19 and -0.92148 values between -0.92 and -0.64379 values between -0.64 and -0.37978 values between -0.37 and -0.091714 values between -0.09 and 0.18258 values between 0.18 and 0.46104 values between 0.46 and 0.7457 values between 0.74 and 1.0140 values between 1.01 and 1.2920 values between 1.29 and 1.5624 values between 1.56 and 1.8416 values between 1.84 and 2.1110 values between 2.11 and 2.3914 values between 2.39 and 2.666 values between 2.66 and 2.944 values between 2.94 and 3.212 values between 3.21 and 3.492 values between 3.49 and 3.762 values between 3.76 and 4.042 values between 4.04 and 4.313 values between 4.31 and 4.594 values between 4.86 and 5.142 values between 5.14 and 5.422 values between 5.42 and 5.691 values between 5.69 and 5.971 values between 7.34 and 7.62offnegativepositiveoften

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. ably
  2. rylic
  3. izable
  4. ized
  5. izing
  6. zed
  7. 谚
  8. Rule
  9. 他又
  10. aving

Suppresses

  1. başına
  2. ниц
  3. .getColumn
  4. antics
  5. /categories
  6. .getInfo
  7. ilians
  8. া�
  9. GetString
  10. -dem