Neuron Explorer

Qwen3-0.6B

'of' after 'all' in phrases

A neuron in layer 16 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 'of' when it comes right after 'all', as in 'all of the students' or 'all of your files'. Its weakest responses are on decimal points and number ranges. Turning it up made the text repeat words like 'good' and 'bad', so it does not clearly change meaning.

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

    keep on your desktop. And all the files get the “.mind” extension added to their filenames. Thus, the infection will not block you from accessing all of your files. However, it can do a lot of other things that can be

  2. #2 peak

    leptin.⏎In their experiment, researchers fed a group of genetically identical mice either a high-calorie diet rich in fats or a low-calorie diet. All of the mice on the low-calorie diet remained lean.⏎Among those that received the

  3. #3 peak

    , they’re so rare that Peter Ackerman wrote a children’s book, “The Lonely Phone Booth,” about one of only four remaining outdoor phone booths in all of Manhattan.⏎“I walked past this phone booth every day with my kid when he was

  4. #4 peak

    committed to providing universal access to all of our virtual events. Please contact email@example.com to request disability accommodations. Advance notice

  5. #5 peak

    't remember the pronunciation, tell them to raise their hands and ask you to say it for them and then have them repeat.⏎- I usually start by having all of the students count from 1-31 with me.⏎- Then I have

  6. #6 peak

    will keep them in their new position until bones grow around the teeth to hold them in place permanently.⏎What Do Braces Look Like?⏎One wire goes across all of your top teeth and another goes across your bottom teeth. It is held in place

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: of the, or of the, of, of, of, of, of? The of, or of the of? The of, or of? The of or?? The of, or?? The of??

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

    Business District as seen from the top of the Carlton Centre⏎|Municipality||City of Johannesburg|⏎|• Total||1.09 km2 (0.42 sq mi)|⏎|• Density||13,000

  2. #2 peak

    vory Silk"⏎Japanese tree lilac (Syringa reticulata) "Ivory Silk" is a 20- to 25-foot tree. Cascading clusters of creamy white florets nearly obscure its long

  3. #3 peak

    feeding habits⏎They feed on crustaceans and small fish.⏎Breeds from June to August, laying planktonic eggs 0.04-0.06 inches (1-1.4 mm) in diameter.⏎No commercial

  4. #4 peak

    and the Japan Aerospace Exploration Agency (JAXA), today announced YouTube Space Lab, a worldwide initiative which extends to the Middle East that challenges 14-18 year-old students to design a science experiment

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. L11:31 suppresses
  2. L13:132 excites "of" after a noun
  3. L15:381 suppresses
  4. L11:18 excites
  5. L1:571 suppresses
  6. L15:286 excites
  7. L2:49 excites
  8. L9:7 excites
  9. L1:87 weakly excites
  10. L14:303 weakly excites
  11. L14:453 weakly suppresses
  12. L5:9 weakly suppresses
  13. L9:30 weakly excites
  14. L15:56 weakly suppresses
  15. L10:372 weakly suppresses
  16. L7:191 weakly suppresses
  17. L1:951 weakly suppresses
  18. L7:102 weakly excites
  19. L5:46 weakly suppresses
  20. L10:33 weakly suppresses
  21. L9:61 weakly excites
  22. L6:4 weakly suppresses
  23. L8:7 weakly suppresses
  24. L5:50 weakly suppresses

Feeds later layers

  1. L19:277 suppresses
  2. L19:217 suppresses
  3. L19:233 suppresses
  4. L19:197 suppresses
  5. L18:1952 excites
  6. L19:283 weakly suppresses
  7. L18:2497 weakly excites hyphen inside compound words
  8. L18:238 weakly excites
  9. L19:287 weakly suppresses
  10. L19:230 weakly suppresses
  11. L19:227 weakly suppresses
  12. L23:2521 weakly excites
  13. L19:403 weakly suppresses
  14. L24:104 weakly suppresses
  15. L19:216 weakly suppresses
  16. L19:129 weakly suppresses
  17. L20:43 weakly suppresses
  18. L25:1530 weakly excites
  19. L20:169 weakly suppresses
  20. L20:217 weakly suppresses
  21. L17:1039 weakly excites
  22. L19:1432 weakly suppresses
  23. L19:243 weakly suppresses
  24. L19:244 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 -4.20 and -3.951 values between -3.23 and -2.983 values between -2.98 and -2.744 values between -2.74 and -2.508 values between -2.50 and -2.265 values between -2.26 and -2.0212 values between -2.02 and -1.7716 values between -1.77 and -1.5321 values between -1.53 and -1.2928 values between -1.29 and -1.0553 values between -1.05 and -0.81159 values between -0.81 and -0.56387 values between -0.56 and -0.32522 values between -0.32 and -0.08676 values between -0.08 and 0.16454 values between 0.16 and 0.41354 values between 0.41 and 0.65291 values between 0.65 and 0.89241 values between 0.89 and 1.13173 values between 1.13 and 1.38157 values between 1.38 and 1.62104 values between 1.62 and 1.8668 values between 1.86 and 2.1062 values between 2.10 and 2.3435 values between 2.34 and 2.5923 values between 2.59 and 2.8321 values between 2.83 and 3.0713 values between 3.07 and 3.317 values between 3.31 and 3.562 values between 3.56 and 3.802 values between 3.80 and 4.041 values between 4.04 and 4.281 values between 4.28 and 4.522 values between 4.77 and 5.011 values between 5.25 and 5.49offnegativepositiveoften

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. mediately
  2. fois
  3. шед
  4. whole
  5. slightest
  6. elaide
  7. ellidos
  8. 多地
  9. えない
  10. whole

Suppresses

  1. /object
  2. oning
  3. /history
  4. ��
  5. /console
  6. méd
  7. пад
  8. villa
  9. ή
  10. Méd