Neuron Explorer

Qwen3-0.6B

capital 'In' starting a word or name

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 fires on a capital 'In' when it starts a fragment of a longer word or place name, like Inca, Inlet, or Inuit. The lowercase 'in' inside words like 'in-class' tends to push it the other way. The steering results were loosely related to 'line' and 'verge' but not clearly tied to this pattern.

Written by Claude from this neuron's activations and steering outputs. Confidence: low. 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

    a major religious site where a ritual was performed, in the shortening days before the winter solstice, to prevent the sun disappearing. Similar stones were at other Inca sites, but were damaged or destroyed by Spanish conquistadores. The one

  2. #2 peak

    Alaska SeaLife Center on September 30th for treatment, and was estimated to be only 2-4 weeks old.⏎He comes from the endangered Cook Inlet beluga population that numbers around 300 individuals; his pod and

  3. #3 peak

    said.⏎That would be a big step forward in identifying the wreck discovered a few miles off the St. Augustine Inlet in the summer of 2009.⏎One of the hypotheses archae

  4. #4 peak

    Commission Science Panel on Coastal Hazards.⏎The detailed, 43-page document draws on tidal gauge measurements at five points along the eastern coast—at Duck, Oregon Inlet, Beaufort, Wilmington and Southport—as well as projections from the

  5. #5 peak

    sea water. The Arctic is a landscape of mountains, fjords, tundra and beautiful glaciers that spawn crystal-colored icebergs. It is the land of Inuit hunters, polar bears and seals and is rich with mysteries that science is still

  6. #6 peak

    p seal and three arctic fox.⏎A visit to the Arctic Ring of Life is more than a fun hike. It opens our eyes to the life of the Inuit<|endoftext|>

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 way the line person person way the line back out the line back back back back back back back back back back back back back back back back back back back back back back back back back back back back back back back back back back back

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

    TESOL-certified and teach in Caoshi Zhen? Are you interested in teaching English in Hengyang Shi? Check out ITTT’s online and in-class courses, Become certified to Teach English as a Foreign Language and start teaching English

  2. #2 peak

    by Jews and Judaism in exile was not the decimation of our numbers, but the injury to Jewish...⏎Become a Member⏎Read all member content⏎Get exclusive in-depth reports from Israel.⏎Get exclusive in-depth reports from Israel⏎Connect with Israel

  3. #3 peak

    Activities⏎- 5 Appreciative Inquiry Games⏎- Appreciative Inquiry Icebreakers⏎- Types of AI Icebreakers⏎- 5 AI Worksheets (Incl. PDF)⏎- Apprec<|endoftext|>Etymology and usage⏎The tribas,

  4. #4 peak

    son investigations are conducted to determine the cause of a fire and ensure the responsible party is held accountable.⏎Arson investigators are usually retired fire fighters or trained professionals with in-depth knowledge about types of fuel, how each fuel ignites and accelerates,

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. L22:535 excites
  2. L17:706 excites
  3. L16:1807 suppresses 'in' after a motion verb
  4. L19:704 excites
  5. L16:1557 excites 'in' meaning 'in this case' or 'in between'
  6. L14:1625 excites the word 'in' after a noun
  7. L17:303 suppresses
  8. L9:1025 excites
  9. L4:289 excites
  10. L20:1574 excites
  11. L14:1389 suppresses
  12. L21:1935 weakly suppresses
  13. L10:132 weakly excites
  14. L5:490 weakly excites
  15. L6:1004 weakly excites
  16. L16:29 weakly suppresses
  17. L20:2850 weakly excites
  18. L18:1588 weakly excites
  19. L8:3068 weakly suppresses
  20. L14:1469 weakly suppresses
  21. L5:861 weakly suppresses
  22. L6:2548 weakly excites
  23. L11:1158 weakly suppresses
  24. L0:705 weakly suppresses 'of' after 'pair' or a landmark noun

Feeds later layers

  1. L26:1865 suppresses
  2. L27:1074 weakly suppresses
  3. L27:2584 weakly excites
  4. L27:1219 weakly excites
  5. L25:2123 weakly excites
  6. L26:1524 weakly excites
  7. L25:2663 weakly excites
  8. L27:721 weakly excites
  9. L25:2029 weakly excites
  10. L26:3018 weakly excites
  11. L25:2225 weakly excites
  12. L25:600 weakly excites
  13. L26:657 weakly suppresses the word 'people' coming next
  14. L26:2255 weakly suppresses
  15. L27:533 weakly suppresses
  16. L26:848 weakly excites
  17. L25:2949 weakly excites
  18. L26:647 weakly excites
  19. L27:177 weakly suppresses
  20. L27:808 weakly suppresses
  21. L25:617 weakly suppresses
  22. L26:2238 weakly suppresses
  23. L27:1949 weakly excites
  24. L25:912 weakly excites the word 'list' and list-like nouns

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 -18.98 and -18.032 values between -14.23 and -13.284 values between -11.38 and -10.434 values between -10.43 and -9.481 values between -9.48 and -8.531 values between -8.53 and -7.583 values between -7.58 and -6.623 values between -6.62 and -5.676 values between -5.67 and -4.726 values between -4.72 and -3.779 values between -3.77 and -2.8223 values between -2.82 and -1.8789 values between -1.87 and -0.922597 values between -0.92 and 0.031032 values between 0.03 and 0.9850 values between 0.98 and 1.9322 values between 1.93 and 2.8813 values between 2.88 and 3.8314 values between 3.83 and 4.795 values between 4.79 and 5.745 values between 5.74 and 6.694 values between 6.69 and 7.643 values between 7.64 and 8.592 values between 8.59 and 9.542 values between 9.54 and 10.491 values between 10.49 and 11.443 values between 11.44 and 12.391 values between 14.29 and 15.241 values between 17.15 and 18.101 values between 18.10 and 19.05offnegativepositiveoften

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. on
  2. ons
  3. ON
  4. onic
  5. /exec
  6. Dam
  7. 一条龙
  8. icone
  9. ivariate
  10. oned

Suppresses

  1. 线
  2. line
  3. 線
  4. line
  5. LINE
  6. (line
  7. Line
  8. -line
  9. box
  10. 规定