Neuron Explorer

Qwen3-0.6B

'two' after 'divided into' or 'there are'

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 reacts to the word 'two' when it introduces a count of things, like 'there are two kinds' or 'divided into two categories'. Turning it up made the model write about counting and lists of items, though the match is loose.

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

    ventilation set to recirculate.⏎If you are very sensitive to smoke,<|endoftext|>What is our proof?⏎The science behind Structured Democratic Dialogues⏎There are two realms for the work of Structured Democratic Dialogues (SDD):⏎The supporting

  2. #2 peak

    ensure that water and air quality are adequately protected during surface and subsurface natural gas development activities.⏎The recent increases in domestic natural gas supplies have been made possible by two technologies—horizontal drilling and hydraulic fracturing—that allow energy companies to tap natural gas

  3. #3 peak

    the power factor, and Realize overcurrent, overvoltage, overload protection and other functions. The main circuit of the frequency converter can be roughly divided into two categories: one is the voltage type, which is a frequency converter that converts the

  4. #4 peak

    stronger footing in terms of offering competitive products to its customers and provide customer satisfaction.⏎(iii) Innovation Innovation is the introduction of new ideas or methods. There are two kinds of innovation in every business⏎(a) Product Innovation In product innovation a new

  5. #5 peak

    And this is where the Digital Identity story begins.⏎Airbnb’s “Verified ID” service uses offline and online methods to ensure the user’s identity. There are two main features relevant in this scenario.⏎Offline validation is performed on a driver’s license

  6. #6 peak

    ates are converted into glucose (also known as blood sugar) and used by the cells for energy purposes.⏎What are complex carbohydrates?⏎Carbohydrates are divided into two main categories: simple and complex carbohydrates. Simple carbohydrates (monosaccharides

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 two and the three, and the two and and and the three, and and and and and, and and and and and, and, and, and, and, and, and, and, and, and, and, and

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

    ⏎• 13-19 July, 2020 - Helsinki, Finland⏎• 20-26 July 2020 – Istanbul, Turkey⏎• 16-

  2. #2 peak

    20 years||100-170 beats per minute||200 beats per minute|⏎|30 years||95-162 beats per minute||190 beats per minute|⏎|35 years||

  3. #3 peak

    the book||$8.32 used (63% off) $17.25 new (22% off) $21.95 direct from Amazon Amazon page|⏎Setup an account with your affiliations in order

  4. #4 peak

    and outbreaks.⏎A snack food at a school fair held on April 26th transmitted Salmonella to 124 attendees – aged one to 56. Twenty-two of the victims needed hospitalization. As of May 2nd

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. L4:87 suppresses
  2. L3:230 excites
  3. L3:986 excites
  4. L10:993 excites
  5. L3:1129 suppresses
  6. L10:773 suppresses
  7. L10:215 excites
  8. L9:1744 excites
  9. L4:326 suppresses
  10. L10:1141 suppresses zero digit inside a number
  11. L5:2823 suppresses
  12. L3:830 weakly excites
  13. L3:3031 weakly excites
  14. L8:41 weakly suppresses
  15. L4:1964 weakly excites
  16. L7:1703 weakly suppresses
  17. L3:856 weakly suppresses
  18. L8:553 weakly suppresses
  19. L14:575 weakly suppresses
  20. L2:1426 weakly excites
  21. L4:1905 weakly suppresses
  22. L4:450 weakly excites
  23. L10:661 weakly suppresses
  24. L10:1901 weakly suppresses

Feeds later layers

  1. L19:1752 excites
  2. L25:3008 excites
  3. L17:1628 excites
  4. L25:1278 excites
  5. L27:1321 excites
  6. L21:808 excites
  7. L19:2842 excites
  8. L22:801 excites the digit 1 in numbers
  9. L26:1201 excites
  10. L17:2388 excites
  11. L24:272 excites
  12. L23:465 excites
  13. L18:2855 excites
  14. L18:726 excites
  15. L20:2151 excites
  16. L23:843 excites
  17. L20:3049 excites
  18. L22:1472 excites
  19. L20:740 excites
  20. L22:2725 weakly excites
  21. L22:939 weakly excites
  22. L17:963 weakly excites
  23. L21:1069 weakly excites quantity words like 'two' and 'several'
  24. L18:2796 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 -4.24 and -3.291 values between -3.29 and -2.342 values between -2.34 and -1.39763 values between -1.39 and -0.453024 values between -0.45 and 0.5056 values between 0.50 and 1.4521 values between 1.45 and 2.4012 values between 2.40 and 3.359 values between 3.35 and 4.303 values between 4.30 and 5.244 values between 5.24 and 6.191 values between 6.19 and 7.141 values between 7.14 and 8.094 values between 8.09 and 9.041 values between 9.98 and 10.931 values between 10.93 and 11.881 values between 15.67 and 16.621 values between 18.52 and 19.471 values between 25.15 and 26.101 values between 32.74 and 33.69offnegativepositiveoften

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. -sided
  2. -dimensional
  3. quire
  4. rellas
  5. -step
  6. urances
  7. lvl
  8. layers
  9. 个月
  10. idable

Suppresses

  1. dale
  2. tracts
  3. maté
  4. alguém
  5. tract
  6. tractor
  7. tracted
  8. 那些
  9. -breaking
  10. Both