Neuron Explorer

Qwen3-0.6B

'the' in location phrases like 'at the end'

A neuron in layer 20 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 lights up on the word 'the' in phrases that point to a place in something, like 'at the bottom', 'at the end' or 'farthest to the left'. When it was turned up, the model got stuck repeating words like 'end' and 'the end of the end'.

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

    pace. To switch between modelspace and paperspace, view the tabs located at the bottom of the screen. One tab is labeled ‘modelspace’ and the other

  2. #2 peak

    sources in the river.⏎Getting a File from HydroShare⏎· Open up Internet Explorer. To do this, type in “Internet Explorer” into the bar at the bottom left of the screen and click on it. It should open up after clicking

  3. #3 peak

    be compared to a wise man who built his house on the rock……⏎Revelation 3:20 'Behold, I stand at the door and knock; if anyone hears My voice and opens the door, I will

  4. #4 peak

    n, the outline method should be replaced with a better note-taking system. Write points in an organized manner based on space indentation. Place major points farthest to the left. Indent each more specific point farther to the right (level of importance

  5. #5 peak

    Enter button twice at the end of each line, it will become an editing nightmare and when the sorting is finished, it will put all of the blank spaces at the top of the list, leaving the data single-spaced anyway.<|endoftext|>A hydrogen

  6. #6 peak

    or family name.⏎Is surname your last name?⏎In the English-speaking world, a surname is commonly referred to as a last name because it is usually placed at the end of a person’s full name, after any given names.⏎What exactly is

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: In all of all all all all all all all all all all all all all all all all all all all all all all all if we are to all all all all all all all all all all all all if all if all if all

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

    medical and family history and your overall health. They will also ask you about your normal menstrual cycle, when you started menstruating, and what symptoms you experience.⏎From there, you will receive a pelvic exam. Your doctor will examine your uterus,

  2. #2 peak

    on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.⏎Disabilities Awareness Rosemary HardySharon Manson⏎“Kyle is

  3. #3 peak

    now verified their hypothesis about the mechanism for that effect. Knowing this may be an important step toward harnessing hypericin's power for more effective disease treatment."⏎From Cow Pasture to Chemistry Lab⏎The study traces its roots back to the mystery

  4. #4 peak

    (CARE) Consortium. The consortium is a joint effort of the National Collegiate Athletics Association (NCAA) and the U.S. Department of Defense.⏎From among the CARE ARC database, researchers evaluated 127 eligible student athletes who

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. L18:2356 excites
  2. L18:483 excites
  3. L12:2140 excites 'the' after place prepositions
  4. L18:2677 excites
  5. L16:1807 suppresses 'in' after a motion verb
  6. L19:1276 suppresses
  7. L19:2820 suppresses
  8. L16:1597 suppresses
  9. L17:2858 excites
  10. L7:1941 suppresses
  11. L10:262 suppresses
  12. L17:832 suppresses
  13. L9:1025 excites
  14. L19:1400 excites
  15. L13:1695 weakly suppresses
  16. L4:1966 weakly suppresses
  17. L7:2192 weakly excites
  18. L19:2448 weakly suppresses
  19. L16:1168 weakly excites 'to' after verbs of change
  20. L0:7 weakly suppresses
  21. L5:972 weakly suppresses
  22. L4:430 weakly suppresses
  23. L6:998 weakly suppresses
  24. L18:1375 weakly suppresses

Feeds later layers

  1. L25:1660 excites
  2. L23:2245 excites "at" before end, bottom or start
  3. L24:2569 excites
  4. L23:2905 excites
  5. L22:2370 excites
  6. L21:1055 excites
  7. L22:1450 excites
  8. L23:748 excites
  9. L21:1269 excites "the" in "on the other hand"
  10. L27:2013 excites
  11. L22:1470 excites
  12. L24:810 weakly excites
  13. L25:1860 weakly excites
  14. L24:327 weakly excites
  15. L27:2471 weakly excites
  16. L27:2233 weakly excites
  17. L22:938 weakly excites
  18. L22:2977 weakly excites
  19. L24:1107 weakly excites
  20. L21:617 weakly excites
  21. L21:1081 weakly excites
  22. L22:207 weakly excites the prefix 'non' before a word
  23. L25:724 weakly excites
  24. L23:2638 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 -11.52 and -10.341 values between -9.17 and -7.991 values between -6.81 and -5.642 values between -5.64 and -4.467 values between -4.46 and -3.2916 values between -3.29 and -2.1143 values between -2.11 and -0.943180 values between -0.94 and 0.24551 values between 0.24 and 1.4138 values between 1.41 and 2.5923 values between 2.59 and 3.7615 values between 3.76 and 4.947 values between 4.94 and 6.124 values between 6.12 and 7.294 values between 7.29 and 8.471 values between 8.47 and 9.641 values between 9.64 and 10.824 values between 10.82 and 11.992 values between 11.99 and 13.173 values between 13.17 and 14.341 values between 14.34 and 15.521 values between 15.52 and 16.691 values between 20.22 and 21.401 values between 34.33 and 35.50offnegativepositiveoften

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. 边缘
  2. 末
  3. end
  4. middle
  5. edge
  6. front
  7. 深处
  8. 中间
  9. top
  10. center

Suppresses

  1. issions
  2. 我们在
  3. 友们
  4. uated
  5. freund
  6. posium
  7. référence
  8. weighed
  9. ConfigurationException
  10. ifiers