Neuron Explorer

Qwen3-0.6B

'so' in 'so-called' and 'sores'

A neuron in layer 24 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

The neuron fires on the syllable 'so' when it starts a word such as 'so-called', 'sores' or 'so far'. Its opposite side reacts to 'far' in phrases like 'so far away' and 'far-distant', and turning it down made the model repeat 'far'.

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

    lockner Hochalpenstraße. The outskirts of the National Park Hohe Tauern are more intensively used, the landscape there is shaped by traditional farming with so-called "Almen".⏎The German term "Alm" refers to a traditional

  2. #2 peak

    purchased through a buy-an-acre program. It is not uncommon for loggers, oil and gas companies, cattle ranchers, and miners to illegally extract resources from so-called “protected” areas.⏎Traditional forest communities are often the best stewards of

  3. #3 peak

    Biology, understanding of how anthropogenic pollutants impact aquatic organisms (aquatic toxicology) can be seen as a natural extension of early efforts to understand how variations in so-called “natural” environmental factors (e.g., oxygen, salinity) affected

  4. #4 peak

    , and difficult to subordinate. On the eve of much of West Africa’s independence from France in 1961, R J Harrison Church spoke of the so-called Dry Zone, the area running horizontally from southern Mauritania across central Mali and

  5. #5 peak

    associated with most forms of chemical peel. These include excessive burning, infection, scarring, hyper-pigmentation, hypo-pigmentation and the development of cold sores. These risks should be understood by patients before they attempt this procedure.More articles

  6. #6 peak

    Where were you hurt?” the general asked in his high-pitched voice (Patton sounded nothing like George C. Scott).⏎The physically-exhausted and sweat-soaked Kuhl could barely lift his eyes up to the general. His mind and

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: far-fremd far-fremd far-fremd far-fremd far-fremd far-fremf... far-fremf... far-fremf... far-fremf... far-fre... A far-frem

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

    inside them. Other stars are large enough to hold several stars the size of the Sun inside them! Stars only appear small in the nighttime sky because they are so far away.⏎In the long ago, astronomers once thought that stars numbered in the hundreds

  2. #2 peak

    conference with a specialist, rural emergency room physicians are more likely to adjust diagnosis and course of treatment for injured children, research shows.⏎The use of technology to link far-distant practitioners

  3. #3 peak

    ___________.⏎In the scientific name of the white oak, Quercus alba, Quercus is the __________________.⏎Two groups of organisms that are farther from each other on a cladogram share __________ derived traits than groups that are

  4. #4 peak

    called a “blackbody”. It’s just the theoretical maximum that can be radiated. Some surfaces are very close to a blackbody and others are a long way off.⏎Note: I have seen many articles by keen budding science writers who have

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. L22:1046 excites
  2. L22:1886 suppresses
  3. L20:1295 suppresses
  4. L20:1512 suppresses
  5. L23:545 excites
  6. L12:2450 weakly suppresses
  7. L23:592 weakly suppresses
  8. L4:2856 weakly suppresses
  9. L23:555 weakly excites
  10. L12:34 weakly suppresses
  11. L20:1960 weakly suppresses
  12. L5:1266 weakly excites
  13. L8:721 weakly excites
  14. L11:1251 weakly suppresses
  15. L21:2777 weakly excites
  16. L17:1809 weakly excites
  17. L0:1372 weakly excites
  18. L5:859 weakly suppresses
  19. L8:2551 weakly suppresses
  20. L12:1693 weakly excites
  21. L18:2561 weakly suppresses
  22. L4:1409 weakly suppresses
  23. L18:562 weakly excites
  24. L12:2669 weakly suppresses

Feeds later layers

  1. L26:2281 suppresses
  2. L25:1765 weakly suppresses
  3. L27:296 weakly suppresses
  4. L25:504 weakly suppresses
  5. L25:2144 weakly excites
  6. L25:2947 weakly excites the digit 1 starting a year
  7. L27:673 weakly suppresses
  8. L27:1591 weakly suppresses
  9. L27:1472 weakly excites
  10. L25:2913 weakly suppresses
  11. L26:2202 weakly excites
  12. L25:1931 weakly excites
  13. L25:770 weakly suppresses
  14. L26:1099 weakly suppresses
  15. L25:2080 weakly excites
  16. L27:2943 weakly suppresses
  17. L27:1369 weakly suppresses
  18. L27:2240 weakly suppresses
  19. L27:745 weakly suppresses
  20. L27:1541 weakly suppresses
  21. L26:2798 weakly suppresses
  22. L26:413 weakly excites
  23. L26:1548 weakly excites
  24. L25:1209 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 -18.19 and -17.041 values between -17.04 and -15.882 values between -15.88 and -14.731 values between -12.42 and -11.272 values between -11.27 and -10.121 values between -10.12 and -8.971 values between -8.97 and -7.815 values between -6.66 and -5.516 values between -5.51 and -4.3624 values between -4.36 and -3.2031 values between -3.20 and -2.05112 values between -2.05 and -0.903061 values between -0.90 and 0.26562 values between 0.26 and 1.4145 values between 1.41 and 2.5617 values between 2.56 and 3.7110 values between 3.71 and 4.874 values between 4.87 and 6.025 values between 6.02 and 7.173 values between 7.17 and 8.333 values between 8.33 and 9.483 values between 9.48 and 10.634 values between 10.63 and 11.782 values between 11.78 and 12.941 values between 12.94 and 14.091 values between 26.77 and 27.92offnegativepositiveoften

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. düzen
  2. eventdata
  3. accumulator
  4. ompson
  5. ồn
  6. obox
  7. plusplus
  8. nil
  9. ingroup
  10. solidity

Suppresses

  1. far
  2. far
  3. 远
  4. Far
  5. Far
  6. 遠
  7. _far
  8. FAR
  9. remote
  10. remote