Neuron Explorer

Qwen3-0.6B

first half of institution names

A neuron in layer 18 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on the middle of organisation names, such as 'U.S. Department', 'National Center' or 'University' of some place, where the rest of the name is still to come. It goes quiet or negative on 'of' in phrases like 'parts of speech'. Turning it up did not change the writing in an obvious, consistent 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

    find the observation such that a quarter of the observations are smaller than it and three quarters are bigger. That value is the lower quartile. Next find the observation such that three quarters of the observations are smaller than it and a quarter are bigger.

  2. #2 peak

    vice-chancellor, probably overwhelmed by the occasion, asked the minister of state for petroleum and natural gas and the minister of state for external affairs to help the university procure an Army tank for the campus

  3. #3 peak

    strengthen the worldwide movement for human rights. (Courtesy of Philip Graitcer)⏎Houck’s Civil Rights Tour of Atlanta begins at the city’s new National Center for Civil and Human Rights.⏎He explains that when he moved here in 1

  4. #4 peak

    . The pipe was off center and buckled quickly.⏎Similar tests were conducted simultaneously at the University of Illinois at Urbana-Champaign, Oregon State University and the University of California at San Diego. The tests were part of the grand opening of the

  5. #5 peak

    helmet use and cooperation from all drivers will reduce fatalities and injuries on our nation’s highways⏎WASHINGTON – To kick off Motorcycle Safety Awareness Month, the U.S. Department of Transportation's National Highway Traffic Safety Administration (NHTSA) is reminding all

  6. #6 peak

    important addition to our understanding of early human development,” says Linzhao Cheng, Ph.D., an associate professor of gynecology and obstetrics and co-director of the stem cell program of the Johns Hopkins Institute for Cell Engineering. “This

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 question of the future of the U.S. and the world in the context of the global climate in the context of the climate- driven by the climate- driven by the climate- driven by the climate- ... ... ... ... ... ...

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

    students practice in identifying and working with the parts of speech to improve reading and writing skills. The CD is divided into six sections, each focusing on a different part of speech: nouns/pronouns, verbs, adjectives, adverbs, conjunction

  2. #2 peak

    Some students will find it difficult if they haven't some familiarity with grammar it moves rather quickly. The first of the three sections (like the books) covers parts of speech; the second deals with the relative clause and sentence combining by addition; and

  3. #3 peak

    is able to deliver a balanced and accurate finding as to the cause of death.⏎Some body tissue and organs require specialised examination. This varies, depending upon the cause of death. In most cases, small samples of tissue are taken for further analysis.

  4. #4 peak

    prepare Cash Budget⏎• Understand and apply Profit Planning concepts to prepare Pro-forma Statement⏎Capital Budgeting & Investment Analysis⏎• Understand and apply time value of money concepts to evaluate different streams of cash flow⏎• Describe the characteristics of capital

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. L16:415 suppresses
  2. L14:2725 suppresses
  3. L14:539 excites
  4. L16:412 suppresses
  5. L14:1353 suppresses
  6. L15:396 suppresses
  7. L14:419 suppresses
  8. L16:1375 suppresses
  9. L17:186 excites
  10. L16:2533 suppresses
  11. L16:1130 suppresses
  12. L17:558 suppresses
  13. L3:68 excites
  14. L14:1558 suppresses
  15. L16:302 excites
  16. L13:1376 excites
  17. L4:154 suppresses
  18. L16:2937 suppresses
  19. L11:940 excites
  20. L5:1921 suppresses
  21. L17:870 suppresses
  22. L13:210 suppresses
  23. L9:29 excites opening parenthesis before an acronym
  24. L15:851 excites

Feeds later layers

  1. L19:315 weakly excites
  2. L20:285 weakly excites
  3. L19:451 weakly suppresses
  4. L21:122 weakly excites
  5. L19:333 weakly suppresses
  6. L19:107 weakly suppresses
  7. L23:2545 weakly excites
  8. L24:147 weakly excites
  9. L26:524 weakly excites
  10. L20:119 weakly excites
  11. L25:1370 weakly excites
  12. L21:2069 weakly suppresses
  13. L19:734 weakly excites
  14. L24:2994 weakly suppresses
  15. L24:2273 weakly suppresses
  16. L20:129 weakly excites
  17. L25:1838 weakly excites
  18. L23:2430 weakly excites
  19. L27:1457 weakly suppresses
  20. L19:416 weakly suppresses
  21. L20:2552 weakly excites
  22. L25:2458 weakly excites
  23. L20:452 weakly suppresses
  24. L25:242 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 -7.19 and -6.781 values between -6.78 and -6.382 values between -6.38 and -5.982 values between -5.98 and -5.574 values between -5.57 and -5.174 values between -5.17 and -4.768 values between -4.76 and -4.368 values between -4.36 and -3.969 values between -3.96 and -3.557 values between -3.55 and -3.1510 values between -3.15 and -2.7513 values between -2.75 and -2.3429 values between -2.34 and -1.9439 values between -1.94 and -1.5365 values between -1.53 and -1.13275 values between -1.13 and -0.73471 values between -0.73 and -0.32645 values between -0.32 and 0.08567 values between 0.08 and 0.49471 values between 0.49 and 0.89348 values between 0.89 and 1.30259 values between 1.30 and 1.70193 values between 1.70 and 2.10133 values between 2.10 and 2.51109 values between 2.51 and 2.9173 values between 2.91 and 3.3146 values between 3.31 and 3.7237 values between 3.72 and 4.1227 values between 4.12 and 4.5314 values between 4.53 and 4.9317 values between 4.93 and 5.332 values between 5.33 and 5.742 values between 5.74 and 6.148 values between 6.14 and 6.551 values between 6.55 and 6.952 values between 6.95 and 7.353 values between 7.35 and 7.761 values between 7.76 and 8.162 values between 8.57 and 8.97offnegativepositiveoften

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. 叹
  2. зв
  3. apache
  4. Apache
  5. adb
  6. الخل
  7. ав
  8. apache
  9. asti
  10. olest

Suppresses

  1. 个乡镇
  2. 旅游局
  3. <<"
  4. omon
  5. <<"
  6. roph
  7. +⏎⏎
  8. →⏎⏎
  9. gard
  10. -->⏎⏎⏎