Neuron Explorer

Qwen3-0.6B

digits inside population and census numbers

A neuron in layer 11 of Qwen3-0.6B. Its strongest snippets are fairly consistent: several of them peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the last digits of large comma-formatted numbers, especially population counts and census figures like '1,526,006'. The negative side picks up odd numeric or table-style values. Turning it up produced rambling text that did not mention numbers in any clear way.

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

    As of the 2010 Census, the city had a population of 1,526,006, growing to 1,547,607 in 2

  2. #2 peak

    in 2002, the 26th largest US city. Its estimated 2002 population was 560,415, but its metropolitan area (including Boulder and Greeley) exceeded 2,

  3. #3 peak

    Virginia's population is estimated at 1,844,128, showing a decline from the 1.853 million recorded at the

  4. #4 peak

    state of Utah. As of the 2010 census, the population was 231,236, making it Utah's fourth-most populous county. Its county seat and largest city

  5. #5 peak

    sites in the body.⏎MET proto-oncogene (GeneID: 4233) has a total length of 125,982 bp, and it is located in the 7q31 locus of chromosome

  6. #6 peak

    ⏎1787, and named in honor of Joseph Addison, the celebrated English writer.⏎Capital, Middlebury. Population, 26,549.⏎is the chief town. This county is bounded on the north by Chitt

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: It's a new year, so it's a good time to measure our progress. So, what it means to do it means the same as the number of people in the room. We can measure it. It's a good time to

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 environment. Organic fabrics, unlike industrial cotton does not release any toxic fumes and are environment friendly.<|endoftext|>Results for 'rail'⏎- Length: 4 characters⏎- Syllables: 1⏎Results for word: rail⏎-

  2. #2 peak

    choice. You’d be amazed how coherent even a very young child’s story is. When he was 3, my son dictated a three page/ 3 line story with a complete story arc about a bubble to me. He drew the pictures

  3. #3 peak

    64⏎BJU Science Grade 2 Activity Manual Teacher's Edition (Third Edition)BJU Press / 2010 / Other$31.6

  4. #4 peak

    2002 population was 560,415, but its metropolitan area (including Boulder and Greeley) exceeded 2,417,908 or more than half the state's population in 19

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. L10:2358 excites
  2. L10:651 weakly suppresses
  3. L10:1178 weakly excites
  4. L8:7 weakly excites
  5. L8:978 weakly excites
  6. L10:1269 weakly suppresses
  7. L10:50 weakly excites
  8. L10:736 weakly excites
  9. L0:493 weakly excites
  10. L8:2503 weakly excites
  11. L4:2563 weakly suppresses
  12. L10:1913 weakly excites
  13. L2:1004 weakly excites
  14. L10:695 weakly excites
  15. L10:773 weakly suppresses
  16. L7:3017 weakly suppresses
  17. L2:2822 weakly suppresses
  18. L5:323 weakly excites
  19. L6:1125 weakly suppresses
  20. L9:1422 weakly suppresses
  21. L7:853 weakly excites
  22. L9:1169 weakly suppresses
  23. L8:3 weakly excites
  24. L3:856 weakly suppresses

Feeds later layers

  1. L12:2434 suppresses
  2. L13:1301 suppresses
  3. L14:929 suppresses
  4. L12:947 suppresses
  5. L14:1809 suppresses
  6. L13:2314 excites
  7. L13:2265 excites
  8. L13:1530 suppresses
  9. L12:455 suppresses
  10. L12:1194 excites
  11. L12:1272 excites
  12. L17:2811 suppresses
  13. L12:1935 suppresses
  14. L13:2333 excites
  15. L12:2679 excites
  16. L19:2011 suppresses
  17. L15:2607 weakly suppresses
  18. L14:2257 weakly suppresses
  19. L18:1571 weakly suppresses
  20. L21:2419 weakly suppresses
  21. L14:536 weakly excites
  22. L13:791 weakly suppresses
  23. L12:1407 weakly excites
  24. L14:253 weakly excites

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 -1.61 and -1.491 values between -1.49 and -1.382 values between -1.38 and -1.262 values between -1.14 and -1.023 values between -1.02 and -0.911 values between -0.91 and -0.792 values between -0.79 and -0.674 values between -0.67 and -0.569 values between -0.56 and -0.4435 values between -0.44 and -0.32181 values between -0.32 and -0.21696 values between -0.21 and -0.091501 values between -0.09 and 0.031041 values between 0.03 and 0.15318 values between 0.15 and 0.2671 values between 0.26 and 0.3819 values between 0.38 and 0.502 values between 0.50 and 0.613 values between 0.61 and 0.732 values between 0.73 and 0.851 values between 0.85 and 0.962 values between 0.96 and 1.082 values between 1.08 and 1.201 values between 1.20 and 1.321 values between 1.43 and 1.552 values between 1.67 and 1.781 values between 1.78 and 1.901 values between 2.60 and 2.722 values between 2.84 and 2.951 values between 2.95 and 3.07offnegativepositiveoften

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. ADATA
  2. Clintons
  3. |i
  4. icense
  5. _batches
  6. ürn
  7. TOT
  8. 더라
  9. INGTON
  10. oti

Suppresses

  1. single
  2. ้าย
  3. 祚
  4. емся
  5. single
  6. imentary
  7. (single
  8. 限度
  9. alone
  10. starter