Neuron Explorer

Qwen3-0.6B

"first" in role phrases

A neuron in layer 13 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 "first" when it names someone's role or rank, such as "its first president" or "the first Black student". Turning it up gave wandering, repetitive text that did not clearly show this pattern.

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

    services. Originally called the Little Rock Congregation, the group formally incorporated as B’nai Israel in 1867 and elected Morris Navra as its first president. Samuel Peck joined the congregation as a spiritual leader the following year.

  2. #2 peak

    in 1914, the⏎Amalgamated Clothing Workers of America was created. At age 26,⏎Hillman was elected its first president, an office he would hold until⏎his death 32 years later

  3. #3 peak

    be identical to their constituents but it matters that they share common experiences and perspectives.⏎This is the case for Jamila Debbech Ksiksi, the first black Tunisian MP. She has always felt that she was in Parliament to defend

  4. #4 peak

    Montpelier, Vermont, on December 26, 1837. A hero of the Civil War and Spanish-American War, he was the first, and to date only, person promoted to Admiral of the Navy.⏎Marihu

  5. #5 peak

    island of Sumatra on Wednesday, triggering a tsunami alert for the Indian Ocean.⏎The quake struck about 430 kilometers from Banda Aceh, the capital of Indonesia's Aceh province, and had a magnitude of 8.7

  6. #6 peak

    ission of Slaves, and Protecting Such of Them as Have Been, or May be Liberated or the New York Manumission Society, and became its first president in 1785.⏎The organization was originally composed of Jay and

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 100 years of 1970s to 2013 to 2015 to 2016 to 2013 to 2014 to 201

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

    21st-century America. But to accomplish this requires an understanding of where these divisions originated.⏎“A Dreadful Deceit” uses the life stories of six people to illustrate the history of race in the

  2. #2 peak

    -ROM/DVD and other supplementary materials are not included as part of eBook file.<|endoftext|>is a former municipality in Akershus⏎, which lends its name to a municipality and a county in Norway⏎. The name originally belonged to a farm which

  3. #3 peak

    at the truth. To find and avoid science’s seductive pitfalls, researchers are studying science itself, a field called meta-research.⏎Now, a paper by three Stanford researchers that delves into those pitfalls appears in Proceedings of the National Academy of

  4. #4 peak

    AttachmentsAccessible Version Accessible version of the tutorial content in PDF format⏎Source and Access Information⏎Aligned StandardsThis vetted resource aligns to concepts or skills in these benchmarks.⏎In this interactive tutorial, learn about the major factors that led to America

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. L11:1055 excites
  2. L12:1301 excites
  3. L12:1941 excites
  4. L12:444 excites
  5. L10:1468 suppresses
  6. L12:493 weakly excites
  7. L12:2220 weakly excites
  8. L12:1705 weakly excites
  9. L12:910 weakly excites
  10. L7:1995 weakly excites
  11. L11:2086 weakly excites
  12. L11:3062 weakly suppresses
  13. L12:922 weakly excites
  14. L11:2566 weakly suppresses
  15. L8:1259 weakly excites
  16. L8:107 weakly excites
  17. L12:2283 weakly excites
  18. L11:2087 weakly suppresses
  19. L11:1040 weakly suppresses
  20. L3:3021 weakly excites
  21. L12:2641 weakly excites
  22. L11:1694 weakly suppresses
  23. L8:1124 weakly excites
  24. L6:1285 weakly excites

Feeds later layers

  1. L14:652 suppresses
  2. L16:1369 excites
  3. L15:632 excites
  4. L14:2528 excites
  5. L14:2597 excites
  6. L20:1131 excites
  7. L17:265 suppresses
  8. L17:2764 excites
  9. L17:1543 weakly excites
  10. L15:414 weakly suppresses
  11. L14:1685 weakly suppresses
  12. L16:2009 weakly excites
  13. L18:2604 weakly excites
  14. L15:851 weakly excites
  15. L15:261 weakly suppresses
  16. L16:619 weakly suppresses
  17. L17:2907 weakly excites
  18. L16:2233 weakly excites
  19. L20:661 weakly excites
  20. L19:2471 weakly suppresses
  21. L18:2160 weakly excites
  22. L15:2148 weakly excites
  23. L16:2365 weakly excites
  24. L15:2720 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 -3.12 and -2.992 values between -2.33 and -2.203 values between -2.06 and -1.933 values between -1.93 and -1.803 values between -1.80 and -1.674 values between -1.67 and -1.542 values between -1.54 and -1.409 values between -1.40 and -1.2716 values between -1.27 and -1.1416 values between -1.14 and -1.0117 values between -1.01 and -0.8829 values between -0.88 and -0.7423 values between -0.74 and -0.6137 values between -0.61 and -0.4887 values between -0.48 and -0.35152 values between -0.35 and -0.21432 values between -0.21 and -0.081837 values between -0.08 and 0.05619 values between 0.05 and 0.18270 values between 0.18 and 0.31121 values between 0.31 and 0.4582 values between 0.45 and 0.5845 values between 0.58 and 0.7138 values between 0.71 and 0.8423 values between 0.84 and 0.9715 values between 0.97 and 1.115 values between 1.11 and 1.249 values between 1.24 and 1.373 values between 1.50 and 1.631 values between 1.63 and 1.771 values between 1.77 and 1.901 values between 1.90 and 2.032 values between 2.03 and 2.16offnegativepositiveoften

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. GPC
  2. gard
  3. xies
  4. g
  5. gar
  6. ierge
  7. 痿
  8. eigenen
  9. these
  10. gis

Suppresses

  1. ernes
  2. 面积约
  3. setSupportActionBar
  4. imest
  5. isto
  6. '"
  7. 两家
  8. 不信
  9. 两天
  10. 三人