Neuron Explorer

Qwen3-0.6B

'on' after a noun like 'impact on'

A neuron in layer 3 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 responds to the word "on" when it follows a noun and introduces what that noun affects, such as "the impact on" or "the effect on". Turning it up gave repetitive, unrelated text.

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

    biological materials, specifically their ability to provide multiple functions despite severe limitations in the quality of available building blocks<|endoftext|>The OED “is widely regarded as the accepted authority on the English language. It is an unsurpassed guide to the meaning, history,

  2. #2 peak

    1961 (called the Principal Act) which⏎received the assent of the President on the 15th April 1962 and was published in the

  3. #3 peak

    natural disasters are essential to reduce loss of life and property by reducing the impact on populations both within the U.S. and abroad. The Federal Emergency Management Agency (

  4. #4 peak

    the Americas, especially in the movement of tobacco and sugar. The industries of Scotland produced cotton, coal and iron which were exported. Shipbuilding became a major industry on the Clyde, building many famous ships. By the end of the 19

  5. #5 peak

    More judicious prescribing of antibiotics for childhood respiratory infection has not increased the number of episodes of peritonsillar abscess or rheumatic fever. The effect on mastoidectomy is unclear, but a clinically important increase seems unlikely. (Level

  6. #6 peak

    and underscored the critical need for effective cooperation on border biosecurity measures.⏎Biosecurity is critical to protecting public health, ecosystems 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: it is a question that is a question that is a Which one is correct? A) the A) the A) the A) the A) the A) the A) the 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

    to a judge from the Salem Witch Trials, which was his Great-Great Grandfather John Hathorne. Hathorne was the only judge who did not express atonement for his crimes,

  2. #2 peak

    made atonement for the dead, so that they might be set free from their sin.”

  3. #3 peak

    It was declared, with a sigh, that most Indians would be removed due merely to the<|endoftext|>Editorial: Team effort needed to combat cyberbullying⏎Updated On: Sep 21 2012 11:49

  4. #4 peak

    .⏎The Hasinai group of the Caddo tribe built a village in the area about

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. L1:1104 suppresses
  2. L1:906 suppresses
  3. L1:1194 suppresses
  4. L1:1 suppresses
  5. L2:1669 weakly excites
  6. L0:1280 weakly excites
  7. L2:1323 weakly suppresses unclear: fragments of rare nouns
  8. L1:2567 weakly excites
  9. L1:2296 weakly excites
  10. L1:2740 weakly suppresses word fragments inside non-English words
  11. L2:1942 weakly suppresses
  12. L2:2615 weakly suppresses
  13. L2:200 weakly suppresses
  14. L1:1179 weakly suppresses 'in' after 'result' (results in)
  15. L2:2552 weakly suppresses
  16. L2:691 weakly excites
  17. L2:2567 weakly excites
  18. L1:2014 weakly excites
  19. L2:2307 weakly excites
  20. L2:297 weakly suppresses
  21. L2:750 weakly excites
  22. L0:1660 weakly suppresses
  23. L1:2439 weakly excites
  24. L2:1215 weakly excites

Feeds later layers

  1. L17:303 excites
  2. L10:132 excites
  3. L17:385 excites
  4. L5:763 excites
  5. L18:293 excites
  6. L20:283 excites
  7. L19:1464 excites
  8. L17:2619 excites
  9. L4:16 suppresses
  10. L4:4 excites
  11. L16:2666 excites
  12. L27:168 excites
  13. L17:3068 excites comma after a sentence subject
  14. L13:132 excites "of" after a noun
  15. L6:613 excites
  16. L16:2120 excites
  17. L17:2302 excites 'in' after a noun
  18. L18:2908 suppresses
  19. L18:2887 excites
  20. L6:846 excites
  21. L22:489 excites
  22. L6:2379 weakly excites
  23. L4:1500 weakly excites
  24. L23:1185 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 -1.73 and -1.612 values between -1.38 and -1.271 values between -0.92 and -0.812 values between -0.81 and -0.694 values between -0.69 and -0.5824 values between -0.58 and -0.46132 values between -0.46 and -0.35455 values between -0.35 and -0.23918 values between -0.23 and -0.121210 values between -0.12 and -0.00707 values between -0.00 and 0.11194 values between 0.11 and 0.2350 values between 0.23 and 0.3424 values between 0.34 and 0.4619 values between 0.46 and 0.5721 values between 0.57 and 0.6918 values between 0.69 and 0.8024 values between 0.80 and 0.9216 values between 0.92 and 1.0312 values between 1.03 and 1.159 values between 1.15 and 1.2612 values between 1.26 and 1.389 values between 1.38 and 1.497 values between 1.49 and 1.616 values between 1.61 and 1.724 values between 1.72 and 1.845 values between 1.84 and 1.955 values between 1.95 and 2.074 values between 2.07 and 2.183 values between 2.18 and 2.302 values between 2.30 and 2.422 values between 2.42 and 2.532 values between 2.53 and 2.651 values between 2.65 and 2.763 values between 2.76 and 2.88offnegativepositiveoften

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. geb
  2. 的就是
  3. ervisor
  4. 年由
  5. HONE
  6. 任何一个
  7. Mastery
  8. ñana
  9. ⏎⏎⏎
  10. .heroku

Suppresses

  1. andid
  2. eselect
  3. filmer
  4. 带走
  5. .removeAll
  6. vice
  7. saç
  8. VIN
  9. idity
  10. wake