Neuron Explorer

Qwen3-0.6B

number theory terms like 'number' and 'Gauss'

A neuron in layer 9 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 lights up on maths words, especially 'number' in phrases like 'algebraic number theory' and the name Gauss. When it was turned up, the model's writing drifted toward maths problems about multiples and common factors.

Written by Claude from this neuron's activations and steering outputs. Confidence: high. 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

    classically Diophantine geometry), a subfield of algebraic number theory.⏎- The study of the real points of an algebraic variety is the

  2. #2 peak

    als. The quadratic residue Gauss walk exhibits a more complex behavior whose shape is roughly determined by the remainder of modulo

  3. #3 peak

    , takes steps given by the terms of the exponential or quadratic residue Gauss sum modulo , and ends at the value of the Gauss sum. By a famous result of Gauss, if the modulus is a prime number, the two walks always end at the

  4. #4 peak

    with coordinates in an algebraically closed field.⏎- The study of the points of an algebraic variety with coordinates in the field of the rational numbers or in a number field became arithmetic geometry (or more

  5. #5 peak

    by different equations.⏎Algebraic geometry occupies a central place in modern mathematics. The concepts it uses connects it to such diverse fields as complex analysis, topology and number theory. At the start, algebraic geometry was about studying systems of polynomial equations

  6. #6 peak

    ured members. Similarly, our estimate predicts 2.5 members starting from 86. It is easy to check that the sequence doesn’t contain any more numbers below 200 and our estimate predicts 0.07 members after

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 name of this? I need to put this in the document, please, I need to refer to the specific sentence, what is the process of doing this, how the data is collected and used in this document, what the

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

    how to keep them from destroying your landscape.⏎Symptoms Of Oak Spider Mites:⏎Spider mites cause damage by sucking cell contents from the leaves. You most likely will not notice anything odd when m

  2. #2 peak

    Xyleborus glabratus.⏎Adult females construct galleries in the sapwood and inoculate the galleries with a fungus (Ophiostoma sp., vascular wilt pathogen). The adults and larvae feed on fungi and not on the wood of

  3. #3 peak

    . In 1981, Chile was officially declared FMD-free without vaccination, after complying with all international health requirements.⏎Foot-and-mouth disease (FMD) affects mainly swine and related wild ruminants and is characterized by so

  4. #4 peak

    ured at the base, where bees occasionally bypass their pollination task by biting through the flower wall to “rob” the nectaries. Watch too for bees landing on discolored leaves and licking them; this behavior transmits Monilinia

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. L6:1047 excites
  2. L6:737 suppresses
  3. L6:1177 excites
  4. L6:7 suppresses
  5. L6:1608 excites
  6. L6:2420 suppresses
  7. L8:2577 excites
  8. L4:2288 weakly excites
  9. L8:5 weakly suppresses
  10. L7:375 weakly suppresses
  11. L6:531 weakly suppresses
  12. L6:2813 weakly excites
  13. L6:2943 weakly excites
  14. L4:2299 weakly suppresses
  15. L7:2830 weakly suppresses
  16. L8:2943 weakly excites 'computer' and 'desktop' after device words
  17. L5:495 weakly suppresses
  18. L8:3035 weakly suppresses
  19. L8:3028 weakly excites
  20. L7:164 weakly excites
  21. L2:177 weakly excites
  22. L3:2246 weakly suppresses
  23. L1:952 weakly suppresses
  24. L0:2638 weakly suppresses

Feeds later layers

  1. L10:978 excites
  2. L10:2187 excites
  3. L21:1637 excites
  4. L19:1033 excites
  5. L16:1848 excites
  6. L22:2949 excites
  7. L17:2958 excites
  8. L21:984 excites
  9. L20:1583 excites
  10. L11:2205 excites
  11. L24:3009 excites
  12. L22:2867 weakly excites
  13. L10:2571 weakly excites
  14. L20:2044 weakly suppresses
  15. L18:1676 weakly suppresses
  16. L23:2372 weakly excites
  17. L20:1929 weakly excites
  18. L22:2807 weakly excites
  19. L18:2628 weakly excites
  20. L20:1203 weakly excites
  21. L24:2207 weakly excites
  22. L21:2594 weakly suppresses
  23. L19:1775 weakly suppresses
  24. L20:454 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 -4.94 and -4.801 values between -2.03 and -1.891 values between -1.61 and -1.471 values between -1.06 and -0.922 values between -0.92 and -0.782 values between -0.78 and -0.643 values between -0.64 and -0.5017 values between -0.50 and -0.36100 values between -0.36 and -0.22591 values between -0.22 and -0.082158 values between -0.08 and 0.06867 values between 0.06 and 0.19141 values between 0.19 and 0.3318 values between 0.33 and 0.475 values between 0.47 and 0.61offnegativepositiveoften

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. ogenerated
  2. gcd
  3. gatsby
  4. payday
  5. hoe
  6. achievable
  7. Enumeration
  8. steen
  9. 修士
  10. gcd

Suppresses

  1. _backward
  2. sigmoid
  3. iana
  4. spl
  5. 抢救
  6. 崴
  7. _ALIAS
  8. jiang
  9. 熄
  10. 仪器