Neuron Explorer

Qwen3-0.6B

the 'Ar' syllable inside names

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 lights up on the letters "Ar" when they appear in the middle of a name, such as Arista, Arroyo, or Arches. Its negative side fires on single digits inside numbers, such as the 4 in 1545. When it was turned up, the model's text was filled with quote marks.

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

    Life examines the work of extraordinary raconteurs Salvador Novo, Cube Bonifant, Roberto Arlt, Alfonsina Storni, and Mário de Andrade,

  2. #2 peak

    -⏎2. Mariano Arista, El general en gefe del cuerpo de Ejército del Norte á

  3. #3 peak

    California's Guaymas Basin in the⏎"We have a huge problem on our hands," says coordinator Juan⏎Pedro Vela Arreola of the Alianza de Ribereños y⏎Armadores, an

  4. #4 peak

    decades, the church in Brazil has shown itself strongly on the side of the poor. Mentioning great figures like Bishop Helder Câmara and Cardinal Paolo Evaristo Arns, Bernard spoke of the prophetic character of the Brazilian church. It has

  5. #5 peak

    .⏎Puente de Colgante⏎The construction of the Puente Colgante was completed in 1852. The bridge linked the district of Arroceros (the area surrounding the Metropolitan Theater and Manila City Hall) to

  6. #6 peak

    visitors than the Grand Canyon or Yellowstone National Park. By comparison, Utah's five National Parks—Zion, Bryce Canyon, Capitol Reef, Canyonlands, and Arches—had a combined total of 5.3 million visitors in 2

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 a single right " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " "

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

    the Catholic Church⏎The Council of Trent (Latin : Concilium Tridentinum), held between 1545 and 1563 in Trent (or Trento), now in

  2. #2 peak

    III, who was the King of England from 1327 to 1377. Edward, who loved knightly pursuits such as war, joust

  3. #3 peak

    015 The University of North Carolina Press⏎116 South Boundary Street, Chapel Hill, NC 27514-3808⏎How to Order | Make a Gift

  4. #4 peak

    we are printing the element for index i and decrementing the value of i by 1.⏎If you run this program, it will print the below output:⏎5 4 3 2 1⏎We need to use the range function

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. L22:336 excites
  2. L17:1665 suppresses
  3. L21:749 suppresses
  4. L23:1360 suppresses
  5. L16:1922 weakly excites
  6. L13:2336 weakly excites
  7. L13:2850 weakly excites
  8. L21:162 weakly excites
  9. L19:796 weakly suppresses
  10. L22:1656 weakly excites
  11. L19:2194 weakly suppresses first half of hyphenated compound adjectives
  12. L19:3035 weakly excites
  13. L18:75 weakly excites
  14. L18:506 weakly suppresses
  15. L18:2255 weakly suppresses
  16. L20:2674 weakly excites
  17. L19:450 weakly suppresses
  18. L20:1587 weakly excites
  19. L3:2870 weakly suppresses
  20. L21:636 weakly excites
  21. L11:1557 weakly suppresses
  22. L0:1673 weakly suppresses
  23. L21:2761 weakly excites
  24. L19:2805 weakly excites

Feeds later layers

  1. L25:1641 strongly suppresses
  2. L27:661 suppresses
  3. L25:161 suppresses
  4. L26:302 suppresses colon or comma before a quote
  5. L27:263 weakly suppresses
  6. L26:897 weakly suppresses the 'gro' fragment in words
  7. L27:33 weakly excites
  8. L27:315 weakly suppresses
  9. L27:2753 weakly excites
  10. L26:1023 weakly suppresses
  11. L26:363 weakly suppresses
  12. L27:825 weakly suppresses
  13. L26:21 weakly suppresses
  14. L27:1521 weakly excites
  15. L27:1045 weakly suppresses
  16. L27:112 weakly excites
  17. L25:1667 weakly suppresses
  18. L25:109 weakly suppresses
  19. L27:292 weakly suppresses
  20. L25:232 weakly suppresses
  21. L26:605 weakly excites
  22. L26:1314 weakly suppresses
  23. L26:439 weakly excites
  24. L25:2443 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 -30.64 and -29.701 values between -7.21 and -6.271 values between -6.27 and -5.331 values between -5.33 and -4.402 values between -4.40 and -3.4635 values between -1.58 and -0.653796 values between -0.65 and 0.2960 values between 0.29 and 1.235 values between 1.23 and 2.171 values between 2.17 and 3.101 values between 4.04 and 4.983 values between 4.98 and 5.911 values between 5.91 and 6.85offnegativepositiveoften

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. }}],⏎
  2. 选
  3. roach
  4. IRO
  5. 不清楚
  6. 蒟
  7. ucose
  8. cente
  9. 薅
  10. ','".$

Suppresses

  1. "
  2. '
  3. “
  4. ‘
  5. «
  6. ’
  7. ”
  8. "⏎
  9. "]
  10. ing