Neuron Explorer

Qwen3-0.6B

'been' after 'has' or 'have'

A neuron in layer 10 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 'been' in phrases like 'has been developed' or 'have been prepared', where a past participle follows. It was most negative on 'have been' in some sentences. Turning it up produced text about things that have been or are happening over time, so the link is mild.

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

    California Institute of Technology in Pasadena, for NASA’s Science Mission Directorate, Washington D.C. UCLA is responsible for overall Dawn mission science. The Dawn framing cameras have been developed and built under the leadership of the Max Planck Institute for Solar System Research

  2. #2 peak

    that was published in November 2011 and the previous edition of the International Basic Safety Standards which was published in 1996. It has been extensively revised and updated to take account of the latest finding of the United Nations Scientific

  3. #3 peak

    Using this item⏎This item has been provided for private study purposes (such as school projects, family and local history research

  4. #4 peak

    Pakistan⏎||It has been suggested that this article be merged with Blasphemy law#Pakistan. (

  5. #5 peak

    he was able to travel as far as Trebizond in the East, and acquire a collection of valuable reliquaries, manuscripts, and icons. It has been suggested that the Mosaic Icon of Saint Demetrios was gifted to Perotti

  6. #6 peak

    this project because of our expertise in planning and conducting structural dynamic tests on very large structures," Carne says.⏎Sandia's solid mechanics/structural dynamics group has done numerous structural analysis projects on large structures including the I-40 Rio Grande bridge

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 "L" in the "L" of the "L" of the "L" in the "L" of the "L" of the "L" of the "L" of the "L" of the "L

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

    admire but also interactive thanks to educational workshops.<|endoftext|>New human genome research aids understanding of disease⏎In the decade since the human genome was published, scientists have been frustrated by their inability to figure out exactly how variations in genes promote disease.⏎But

  2. #2 peak

    responds to training depends on his breed’s history, genetic makeup, individual personality and, of course, you.⏎Genetics and Dog Behavior⏎Humans have bred dogs for specific duties for thousands of years, to the point where certain behavioral and personality traits

  3. #3 peak

    Eckstorm concludes the best interpretation is “shaped like a great knee,” referring to the peninsula on which much of the city sites.⏎It has had its share of mishaps: destruction in 1676 during King Phillips War, burned

  4. #4 peak

    astonishment in the face of this captured<|endoftext|>The UN adopted the Universal Human Rights Declaration in 1948 and since that time countries and international bodies have been adding to the fundamental rights of people.⏎Human Rights Law is governed by a variety

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. L5:660 strongly suppresses
  2. L6:669 excites
  3. L6:535 excites
  4. L8:2719 suppresses
  5. L5:536 suppresses
  6. L8:255 suppresses
  7. L7:153 suppresses
  8. L8:1626 suppresses
  9. L5:527 suppresses
  10. L9:670 suppresses
  11. L7:734 excites
  12. L4:457 excites
  13. L3:1260 weakly suppresses
  14. L6:2614 weakly excites
  15. L8:3049 weakly excites
  16. L4:324 weakly suppresses
  17. L2:159 weakly excites
  18. L8:800 weakly excites
  19. L9:983 weakly suppresses
  20. L9:2508 weakly excites
  21. L1:2950 weakly excites
  22. L9:2088 weakly excites
  23. L2:834 weakly excites
  24. L4:1404 weakly excites

Feeds later layers

  1. L18:1968 excites
  2. L12:2964 excites
  3. L20:2574 suppresses
  4. L15:2901 suppresses
  5. L11:2896 excites
  6. L12:2898 excites
  7. L18:863 suppresses
  8. L12:1639 excites
  9. L12:1431 excites
  10. L12:742 suppresses
  11. L23:535 suppresses
  12. L11:1669 suppresses
  13. L18:2640 excites
  14. L11:2573 suppresses
  15. L17:2060 suppresses
  16. L11:1360 weakly excites
  17. L14:2939 weakly suppresses
  18. L21:773 weakly suppresses
  19. L16:1532 weakly suppresses
  20. L11:1163 weakly excites
  21. L18:642 weakly excites
  22. L13:2036 weakly suppresses
  23. L18:620 weakly excites
  24. L17:1255 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 -4.68 and -4.451 values between -3.56 and -3.341 values between -3.11 and -2.891 values between -2.00 and -1.779 values between -1.10 and -0.8822 values between -0.88 and -0.66181 values between -0.66 and -0.43748 values between -0.43 and -0.212041 values between -0.21 and 0.01794 values between 0.01 and 0.2462 values between 0.24 and 0.4619 values between 0.46 and 0.686 values between 0.68 and 0.915 values between 0.91 and 1.131 values between 1.13 and 1.351 values between 1.35 and 1.586 values between 1.58 and 1.802 values between 1.80 and 2.021 values between 2.02 and 2.252 values between 2.47 and 2.692 values between 2.69 and 2.911 values between 2.91 and 3.141 values between 4.03 and 4.25offnegativepositiveoften

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. s
  2. enc
  3. Adv
  4. obs
  5. ?>⏎
  6. orts
  7. ?>
  8. 0
  9. 不断提高
  10. orse

Suppresses

  1. 一开始
  2. invaded
  3. ",⏎⏎
  4. 当时的
  5. Started
  6. SenderId
  7. ẏ
  8. invade
  9. 一夜
  10. 一瓶