Neuron Explorer

Qwen3-0.6B

'of' after type or class nouns

A neuron in layer 12 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' in phrases like 'a specific type of', 'one interesting class of' or 'a special class of', where a category is about to be named. It seems to go quiet or negative on 'of' or 'in' after a quantity or a day, such as 'the second Monday of'. Turning it up did not produce a clear theme in the model's output.

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

    one interesting class of question is aimed at finding out the variety of situations those laws allow. For example

  2. #2 peak

    active periods of years separated by decades of quiescence, are called persistent SXTs. In 2000, a special rare type of X-ray superbursts was discovered. Superbursts last for a few to

  3. #3 peak

    not only mathematicians - like to consider more specific questions in order to get some feeling for what is happening. In this article we will reflect about a special class of Lunda-designs.⏎On the day before my daughter Likilisa's fourth

  4. #4 peak

    CeMM Forschungszentrum für Molekulare Medizin der Österreichischen Akademie der Wissenschaften⏎An important step towards a completely new experimental access to quantum physics has been made at University of Konstanz. The

  5. #5 peak

    . Of this number, around 30% will develop ER-negative tumors, making a woman’s risk of developing this specific type of cancer at some point in their lives around 3.72%. If it

  6. #6 peak

    , strep throat can cause complications, such as kidney inflammation or rheumatic fever. Rheumatic fever can lead to painful and inflamed joints, a specific type of rash, or heart valve damage.⏎Strep throat is most common in children,

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: what is the way to go, and I have to do this, and I can't find it, but it seems like it, and I have to do this, and I have to do this, and I can't find it,

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

    residues after fossil fuels are burned and cool power plants.⏎Coal uses up to 3.2 cubic metres of water per megawatt hour (MWh) of electricity produced, gas uses up to 1.7 m3 per MWh

  2. #2 peak

    and Congress passed a bill to establish a federal holiday in 1894. President Grover Cleveland signed the bill soon afterward, designating the first Monday in September as Labor Day.⏎Learners are Leaders, Inc.’s mission includes

  3. #3 peak

    the purpose of taking the vote of the electors of the State, for the acceptance or rejection of this Constitution, an election shall be held on the second Monday of November, in the year 1857, to be conducted according to

  4. #4 peak

    The Cassini-Huygens mission is a cooperative project of NASA, the European Space Agency and the Italian Space Agency. The Jet Propulsion Laboratory, a division of the California Institute of Technology in Pasadena, manages

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. L6:2217 suppresses
  2. L11:2497 suppresses
  3. L5:2364 excites
  4. L9:2566 excites
  5. L10:469 suppresses 'of' after a noun phrase
  6. L11:2953 weakly suppresses
  7. L9:1930 weakly excites
  8. L10:2083 weakly excites
  9. L11:2207 weakly suppresses "of" after a noun in trademark and title phrases
  10. L5:363 weakly excites
  11. L9:799 weakly suppresses
  12. L11:2860 weakly suppresses
  13. L4:1076 weakly excites
  14. L5:2108 weakly suppresses
  15. L8:1965 weakly suppresses
  16. L8:685 weakly excites
  17. L6:2816 weakly excites
  18. L11:1174 weakly suppresses
  19. L9:2766 weakly suppresses
  20. L6:2277 weakly excites
  21. L7:2157 weakly excites 'the' after 'like' or 'stay'
  22. L10:886 weakly suppresses
  23. L10:418 weakly suppresses
  24. L9:2659 weakly excites

Feeds later layers

  1. L14:785 suppresses
  2. L16:1128 excites
  3. L15:1514 excites
  4. L14:1953 suppresses
  5. L15:632 excites
  6. L15:283 suppresses
  7. L18:2958 excites
  8. L15:2455 excites
  9. L18:706 suppresses
  10. L17:663 excites
  11. L18:1502 excites
  12. L16:2886 suppresses
  13. L17:1241 excites
  14. L18:2788 excites
  15. L15:285 excites
  16. L14:1971 excites
  17. L17:1240 excites
  18. L16:3062 suppresses
  19. L18:1358 excites
  20. L16:2705 excites
  21. L13:2663 excites
  22. L17:1915 excites
  23. L18:1605 suppresses
  24. L16:2612 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.

4 values between -2.01 and -1.771 values between -1.77 and -1.533 values between -1.53 and -1.2811 values between -1.28 and -1.0420 values between -1.04 and -0.8051 values between -0.80 and -0.55105 values between -0.55 and -0.31790 values between -0.31 and -0.072578 values between -0.07 and 0.18268 values between 0.18 and 0.4235 values between 0.42 and 0.6619 values between 0.66 and 0.915 values between 0.91 and 1.153 values between 1.15 and 1.394 values between 1.39 and 1.641 values between 1.64 and 1.883 values between 2.12 and 2.371 values between 2.37 and 2.611 values between 2.61 and 2.851 values between 2.85 and 3.101 values between 3.58 and 3.831 values between 3.83 and 4.071 values between 4.56 and 4.801 values between 7.48 and 7.72offnegativepositiveoften

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. thereof
  2. @stop
  3. ㄱ
  4. ame
  5. Ơ
  6. GB
  7. 武功
  8. /people
  9. gba
  10. /games

Suppresses

  1. itives
  2. 有用
  3. riott
  4. xic
  5. zsche
  6. ulet
  7. preh
  8. ITIONS
  9. ечение
  10. ova