Neuron Explorer

Qwen3-0.6B

"the" after "and" in academic prose

A neuron in layer 18 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 "the" when it follows a word like "and" or "increasing" in formal writing, such as "infectious diseases and the mechanisms". Turning it up made the model write stuttering, repeated phrases.

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

    41/ aaajmr-6.11.155⏎Citation: Wook Y. Analyses of the trends of infectious diseases and the mechanisms of Adaptive diseases. Allied J Med

  2. #2 peak

    22) Volume 6, Issue 11⏎Analyses of the trends of infectious diseases and the mechanisms of Adaptive diseases.⏎Department of Pediatrics, Soroka University Medical Center, Beer

  3. #3 peak

    result of this work, those tests became standard care in the routine clinical practice of diagnosing individuals thought to have contracted HIV.⏎A prospective study of HIV infection and the development of AIDS in subjects with hemophilia showed that a much larger proportion of HIV

  4. #4 peak

    access routes to the health facility are open and when the water supply and electric power and telecommunications can continue supplying the health facility, thus guaranteeing continuity of operations and the ability to absorb the additional demand for medical care.⏎When hospital practices elements that can

  5. #5 peak

    ⏎From childhood we tend to assign games a strictly entertaining role. However, there is a close connection between the beginning of an early physical activity of a child and the motor skills that the child develops later. Initially, this connection was only noticed,

  6. #6 peak

    so that maximum capabilities can be globally enforced. Even though the approach is computationally intensive and comparably expensive, it shows promise for the future of electric vehicles and the aerospace industry. This research is supported by the Grainger Center for Electric Machinery 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 the the a lot of the a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a

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

    they are the only ones a naked eye can see. The word itself, planet, comes from the ancient Greek word planets, meaning wanderers. While stars look the same night after night, slowly moving along the year and coming back in the sky

  2. #2 peak

    pretentious: "Anything audacious and ambitious has a tincture of pretentiousness in it. Any less than arrogant composer must sometimes feel, ‘Who the hell am I to set forth this

  3. #3 peak

    Cooper.⏎Buy the hardcover edition of⏎I Heard the Willow Weep for $6.00. (Retail price: $1

  4. #4 peak

    , based on a combination of the individual dog’s training and any circumstances they are in at the time.⏎Great Danes And Their History⏎Don’t let the name fool you. The Great Dane breed comes from German descent without any known associations

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. L17:30 suppresses
  2. L16:419 excites
  3. L4:52 excites
  4. L9:24 excites "the" before a named organisation or title
  5. L16:1320 suppresses
  6. L16:792 excites "the" inside a prepositional phrase
  7. L8:636 suppresses
  8. L14:169 suppresses
  9. L17:985 excites
  10. L12:435 suppresses
  11. L17:372 suppresses
  12. L10:120 suppresses
  13. L6:443 suppresses
  14. L6:1782 excites 'of' after a category noun
  15. L17:1999 suppresses
  16. L6:61 suppresses
  17. L9:662 suppresses
  18. L17:714 suppresses
  19. L6:1684 suppresses
  20. L17:6 excites
  21. L10:1377 suppresses
  22. L5:125 excites "of" after all three, both, ten
  23. L17:507 weakly suppresses 'of' after a noun
  24. L16:1851 weakly suppresses

Feeds later layers

  1. L19:1523 excites
  2. L20:1652 excites
  3. L19:1608 excites
  4. L20:585 suppresses
  5. L19:2432 excites
  6. L19:2456 suppresses
  7. L19:689 excites
  8. L21:614 excites
  9. L20:2454 suppresses
  10. L19:2564 suppresses "be" after a modal verb
  11. L24:153 suppresses
  12. L25:247 suppresses
  13. L20:1912 excites
  14. L20:2993 suppresses
  15. L19:1509 excites
  16. L19:1788 suppresses
  17. L19:1166 suppresses
  18. L19:2815 excites
  19. L19:716 excites
  20. L20:219 excites "the" after time or place prepositions
  21. L26:1941 excites
  22. L20:1727 weakly excites
  23. L19:2123 weakly suppresses
  24. L20:2842 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 -8.12 and -7.491 values between -7.49 and -6.873 values between -6.25 and -5.624 values between -5.62 and -5.002 values between -5.00 and -4.376 values between -4.37 and -3.7510 values between -3.75 and -3.136 values between -3.13 and -2.5015 values between -2.50 and -1.8815 values between -1.88 and -1.25139 values between -1.25 and -0.632152 values between -0.63 and -0.011376 values between -0.01 and 0.6257 values between 0.62 and 1.2425 values between 1.24 and 1.8720 values between 1.87 and 2.4920 values between 2.49 and 3.127 values between 3.12 and 3.749 values between 3.74 and 4.362 values between 4.36 and 4.994 values between 4.99 and 5.616 values between 5.61 and 6.246 values between 6.24 and 6.862 values between 6.86 and 7.481 values between 7.48 and 8.113 values between 8.11 and 8.734 values between 8.73 and 9.363 values between 9.36 and 9.981 values between 9.98 and 10.602 values between 10.60 and 11.231 values between 11.23 and 11.852 values between 11.85 and 12.481 values between 12.48 and 13.102 values between 16.22 and 16.84offnegativepositiveoften

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. "<?
  2. Indented
  3. род
  4. strup
  5. roduction
  6. ضاء
  7. xed
  8. ajo
  9. ernes
  10. jal

Suppresses

  1. 一看
  2. anyl
  3. ıl
  4. the
  5. 越來越
  6. })}⏎
  7. integrate
  8. )V
  9. _merge
  10. 礼仪