Neuron Explorer

Qwen3-0.6B

space before a number, often an age

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 responds to the blank space just before a number, especially when the number is an age or a count of years, like 'at age 22' or 'just 18 and small'. It does not do much else that is clear, and the steering outputs drifted to unrelated topics or to repeated digits.

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

    family than her husband.⏎El Cid was raised in the court of Ferdinand I, where he was promoted to armiger regis (standard bearer) at age 22. He had also receive the title of "The Champion", which in

  2. #2 peak

    ist:⏎Baldwin, born in 1795, moved to Philadelphia from his birthplace of Elizabethtown, New Jersey, at 16 years old. Originally trained as a jeweler, he went on to

  3. #3 peak

    visual aspects but ignore verbal signals, which are more prevalent in the media, especially TV."⏎Taylor interviewed nearly 200 college undergraduates between the ages of 18 to 26. The students were randomly assigned to groups and shown

  4. #4 peak

    the NFB collection (60 in English; 60 in French in the French part). Of those, 48 (24 in English; 24 in French in the French part) include an audio description for the blind

  5. #5 peak

    smith trade before enlisting in the Union Army.⏎In 1861, William enlisted in the Pennsylvania 11th Reserve Infantry. He was just 18 and small in stature. Military officials refused to swear him in without the

  6. #6 peak

    three and a half decades. The average age of first-time mothers has increased by 3.6 years over last 36 years, from the age of 21.4 years in 1970 to the age of

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: I want to get back to one of the two standard scale scale score scale score scale score scale scale scale scale scale scale scale score scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale scale

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 land and its resources is a basic need for effective government and a productive economy in any nation. The U.S. Geological Survey (USGS), celebrating its 13

  2. #2 peak

    teacher will give you a set of cards. Match each scatter plot with a pair of variables. Be prepared to explain your reasoning.<|endoftext|>The Saturday Evening Post turns 200 this year, so we are bringing you excerpts from our archive to

  3. #3 peak

    problems (7 points each, so 35 points). Part C has 3 traditional problems (15 points each so 45 points). Total is 92 points. Closed book, no calculators or computers but you may use

  4. #4 peak

    alleable but harder than calcium. It is chiefly employed (as in the nitrate) to color pyrotechnic flames red. Symbol Sr. Atomic weight 87

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. L23:2592 strongly excites
  2. L21:1732 strongly excites
  3. L23:775 strongly excites
  4. L22:2854 excites
  5. L22:610 excites
  6. L21:2502 excites
  7. L21:1954 suppresses
  8. L23:1774 excites
  9. L23:1858 excites
  10. L22:623 excites
  11. L22:2408 excites
  12. L23:2835 suppresses
  13. L21:2918 suppresses
  14. L23:2142 suppresses
  15. L21:326 suppresses
  16. L23:2882 excites
  17. L12:1094 weakly excites
  18. L22:2146 weakly suppresses
  19. L23:2093 weakly suppresses
  20. L20:780 weakly excites 'of' before a measured value
  21. L20:2125 weakly excites
  22. L23:1255 weakly excites
  23. L21:1104 weakly suppresses
  24. L22:3013 weakly suppresses

Feeds later layers

  1. L25:342 excites
  2. L25:1020 excites
  3. L25:1478 excites
  4. L27:1100 suppresses
  5. L25:1087 suppresses word right before a number value
  6. L27:205 excites
  7. L26:580 weakly suppresses
  8. L25:1172 weakly excites
  9. L27:1157 weakly excites
  10. L25:2383 weakly suppresses
  11. L26:754 weakly excites
  12. L25:2029 weakly suppresses
  13. L25:1613 weakly suppresses
  14. L25:2971 weakly suppresses
  15. L25:845 weakly suppresses
  16. L27:2392 weakly suppresses
  17. L25:2664 weakly suppresses
  18. L26:3036 weakly excites
  19. L25:1702 weakly suppresses
  20. L26:905 weakly suppresses
  21. L26:522 weakly suppresses
  22. L25:1578 weakly excites
  23. L26:868 weakly excites
  24. L26:834 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 -27.84 and -26.801 values between -22.61 and -21.561 values between -19.46 and -18.421 values between -13.18 and -12.131 values between -10.03 and -8.991 values between -8.99 and -7.941 values between -7.94 and -6.892 values between -6.89 and -5.842 values between -5.84 and -4.802 values between -4.80 and -3.757 values between -3.75 and -2.706 values between -2.70 and -1.65250 values between -1.65 and -0.613386 values between -0.61 and 0.44169 values between 0.44 and 1.4927 values between 1.49 and 2.5418 values between 2.54 and 3.595 values between 3.59 and 4.6311 values between 4.63 and 5.682 values between 5.68 and 6.738 values between 6.73 and 7.781 values between 7.78 and 8.822 values between 8.82 and 9.871 values between 9.87 and 10.921 values between 10.92 and 11.971 values between 13.02 and 14.06offnegativepositiveoften

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. achable
  2. ropol
  3. 发现
  4. iotic
  5. 指出
  6. 发现自己
  7. 발견
  8. iti
  9. iter
  10. _UNIQUE

Suppresses

  1. PLY
  2. .backward
  3. готов
  4. backtrack
  5. nation
  6. oriously
  7. bum
  8. osas
  9. .Flag
  10. ders