Neuron Explorer

Qwen3-0.6B

'not' or 'non' inside word fragments

A neuron in layer 23 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 responds to the letters 'not' or 'non' when they appear as a piece of a longer word, such as Nottingham or non-stop. It is switched strongly the other way by 'non' as a prefix, so its meaning is hard to pin down. Turning it up did not produce a clear theme.

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

    20th / 21st-century philosophy|⏎|Main interests||Linguistics · Psychology⏎Philosophy of language⏎Politics · Ethics⏎|Notable ideas||Generative grammar|⏎Avram Noam Chomsky (born

  2. #2 peak

    .⏎Additional Physical Format: Online version: Berkhof, Louis, Principles of Biblical interpretation. Grand Rapids, Baker Book House, Principles of Interpretation As RECOGNIZED GENERALLY BY BIBLICAL SCHOLARS, TREATED AS

  3. #3 peak

    other health care professional.⏎This site is published by T-1 Today, Inc. (d/b/a Children with Diabetes), a 501c3 not-for-profit organization, which is responsible for its contents. Our mission is to provide

  4. #4 peak

    States and Mexico; The opposite occurs with Cubans and Puerto Ricans living in the United States.⏎This is reflected in a survey carried out between Axios and Noticias Telemundo. Cubans have settled very well in Florida since the times

  5. #5 peak

    opposition but succeeded in capturing Bois de Vert, Bois du Sart,⏎and reaching the western outskirts of Haucourt, Remy, Boiry-Notre-Dame⏎Immediately the barrage stopped⏎the infantry advance became more difficult.

  6. #6 peak

    has many causes, including hepatitis, gallstones, and tumors. In adults, jaundice usually does not need to be treated. It is also known as icterus.⏎Jaundice most often happens as a result of an underlying disorder

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: if the product of the five numbers of the set are 12000000, and if the numbers are all are of the form 2, 5, 1, 1, 1, 1

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

    and aggressive anyway?⏎Why are geese so loud?⏎Most people are familiar with the honk of the Canadian goose. Sometimes, it seems like they honk non-stop. Geese make so much noise for several reasons:⏎● To protect their

  2. #2 peak

    to protect the earth’s climate. The global climate crisis is non-negotiable. #Move the Date: We need to do everything we can

  3. #3 peak

    there is high suspicion for CAH (such as ambiguous genitalia), further evaluation is also indicated.⏎Later in Life⏎Newborns do not show symptoms of nonclassic CAH, and the test done on newborns does not detect nonclassic

  4. #4 peak

    the motion carries no legal weight and would be unlikely to give Quebec separatists a lever to extract more political powers from Ottawa.So, indeed, this would be a non-nation, nation. In fact, what it comes down to, as cabinet

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. L20:1414 excites
  2. L19:2870 excites
  3. L19:1335 weakly suppresses
  4. L16:2351 weakly excites
  5. L6:13 weakly excites
  6. L10:646 weakly suppresses
  7. L5:98 weakly suppresses
  8. L16:2340 weakly suppresses
  9. L5:555 weakly suppresses
  10. L21:2507 weakly excites
  11. L0:626 weakly excites
  12. L20:1386 weakly suppresses
  13. L8:6 weakly suppresses
  14. L15:3005 weakly suppresses
  15. L5:706 weakly suppresses the number 1 starting a list
  16. L1:428 weakly excites
  17. L9:1330 weakly excites
  18. L7:2858 weakly excites
  19. L0:763 weakly excites
  20. L3:144 weakly excites
  21. L7:2373 weakly suppresses
  22. L5:45 weakly excites
  23. L21:70 weakly excites
  24. L9:233 weakly excites

Feeds later layers

  1. L26:797 weakly suppresses
  2. L24:319 weakly suppresses
  3. L26:109 weakly excites
  4. L27:804 weakly suppresses
  5. L24:2413 weakly suppresses
  6. L25:2529 weakly excites
  7. L24:1150 weakly suppresses
  8. L24:2021 weakly suppresses
  9. L25:2472 weakly suppresses
  10. L27:2521 weakly suppresses
  11. L27:1540 weakly suppresses
  12. L24:469 weakly suppresses
  13. L27:2626 weakly excites
  14. L27:1951 weakly suppresses
  15. L25:822 weakly suppresses
  16. L24:122 weakly suppresses
  17. L25:2048 weakly excites
  18. L26:2576 weakly suppresses
  19. L27:2302 weakly excites
  20. L27:648 weakly suppresses
  21. L25:753 weakly excites
  22. L26:1078 weakly suppresses
  23. L25:2161 weakly suppresses
  24. L26:2848 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 -54.13 and -52.451 values between -33.99 and -32.312 values between -12.17 and -10.491 values between -8.81 and -7.132 values between -7.13 and -5.454 values between -5.45 and -3.789 values between -3.78 and -2.10243 values between -2.10 and -0.423624 values between -0.42 and 1.2614 values between 1.26 and 2.943 values between 2.94 and 4.623 values between 6.30 and 7.971 values between 11.33 and 13.01offnegativepositiveoften

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. iates
  2. endcode
  3. ound
  4. ordinated
  5. ropriate
  6. aras
  7. illac
  8. MaterialApp
  9. assium
  10. ellites

Suppresses

  1. 是
  2. 因
  3. とも
  4. factor
  5. 由
  6. fic
  7. resist
  8. recur
  9. dispos
  10. 無料