Neuron Explorer

Qwen3-0.6B

comma after a sentence subject

A neuron in layer 21 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 a comma that comes straight after the subject of a sentence, such as 'Nicolas Sarkozy,' or 'Rural girls,', and before the verb comes. Turning it up gave text that was not clearly related to this.

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

    posited answers to this question, whether directly - in the case of Plato and Plotinius - or through story - in the case of Homer.⏎Plotinus, having been a proponent of Platonism, naturally concurs with Plato,

  2. #2 peak

    François Mitterand who, in 1954, defiantly pronounced that ‘L’Algérie, c’est la France’. Nicolas Sarkozy, on the other hand, dramatically distanced himself both from Chirac and from

  3. #3 peak

    Manhattan Project, decided that they would declare an American monopoly over the atom.⏎Logically enough, support for this idea was not universally shared. Great Britain and France, two nations with long histories of nuclear science and technology were determined to make progress in

  4. #4 peak

    four years.⏎The disparity is partly due to the immense impact of rural students (by comparison, the urban poor are in school for over six years). Rural girls, however, get only three years of education, on average.⏎While there are a

  5. #5 peak

    influence of the ducal court and supporters of the powerful bishops of Toul, Metz and Verdun. Numerous artists came from Flanders, Germany and, sometimes, Italy to work for the Dukes of Lorraine and their friends.

  6. #6 peak

    decade and perhaps longer, the growth path and survival rate of new businesses means they are generating fewer new jobs.⏎The businesses that started in 2009, for example, are on course to create 1 million fewer jobs in the next

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: I,,e,.,,e,.,,e,.,,e,.,,e,.,,e,.,,e,.,,e,.,,e,.,,e,.,,e,.,,e

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

    may be happy with a confidence level of 95 percent (giving a margin error of 0.05), or you may require a tighter accuracy of a 98 percent confidence level (a margin of error

  2. #2 peak

    (both images from Fri Aug 9 2019)⏎“Similar particles have been detected near Saturn, suggesting⏎that the process is at work there as well,” says Liu. “Indeed,” adds⏎Angelopoulos, “this type

  3. #3 peak

    upon it) wreaks may not be the only way that complexity grew.⏎Protein flaws called “dehydrons” may have made proteins less stable in water and made them more likely to end up working together, building up complex function.⏎Michael

  4. #4 peak

    which can significantly increase a person’s risk of developing diabetes, he says.⏎However, there is evidence that if you lose weight, you have fewer dying fat cells and significantly fewer fat-engorged macrophages, Green

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. L20:1441 strongly suppresses
  2. L20:82 excites
  3. L19:934 excites
  4. L18:174 excites
  5. L20:663 suppresses
  6. L20:1358 suppresses
  7. L18:2376 excites
  8. L20:2730 excites
  9. L20:2674 suppresses
  10. L18:1791 suppresses
  11. L19:1018 excites
  12. L17:1711 excites
  13. L18:2593 excites
  14. L17:723 suppresses
  15. L19:949 excites
  16. L20:1493 excites
  17. L19:2999 excites
  18. L19:2025 excites comma before 'however'
  19. L19:682 suppresses
  20. L20:2788 excites
  21. L17:1845 excites
  22. L19:845 suppresses
  23. L19:650 suppresses
  24. L19:767 excites

Feeds later layers

  1. L26:236 strongly suppresses
  2. L27:86 strongly excites
  3. L26:22 strongly excites the 'res' fragment inside words
  4. L27:358 strongly suppresses
  5. L22:368 strongly suppresses
  6. L27:130 strongly excites
  7. L25:328 strongly suppresses
  8. L27:191 strongly suppresses
  9. L24:216 strongly suppresses
  10. L25:438 strongly suppresses
  11. L22:158 strongly excites
  12. L25:1221 strongly suppresses
  13. L27:286 strongly suppresses
  14. L26:80 strongly excites
  15. L26:345 strongly suppresses
  16. L24:572 suppresses
  17. L22:940 suppresses
  18. L24:304 suppresses
  19. L27:77 excites
  20. L26:307 excites
  21. L24:32 suppresses
  22. L26:844 excites
  23. L27:139 excites
  24. L24:41 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.

3 values between -8.88 and -8.073 values between -8.07 and -7.268 values between -7.26 and -6.4519 values between -6.45 and -5.6317 values between -5.63 and -4.8232 values between -4.82 and -4.0137 values between -4.01 and -3.2070 values between -3.20 and -2.3897 values between -2.38 and -1.57159 values between -1.57 and -0.76629 values between -0.76 and 0.052394 values between 0.05 and 0.87290 values between 0.87 and 1.6828 values between 1.68 and 2.4924 values between 2.49 and 3.3020 values between 3.30 and 4.1124 values between 4.11 and 4.9315 values between 4.93 and 5.746 values between 5.74 and 6.555 values between 6.55 and 7.369 values between 7.36 and 8.186 values between 8.18 and 8.991 values between 8.99 and 9.802 values between 9.80 and 10.612 values between 10.61 and 11.432 values between 11.43 and 12.242 values between 12.24 and 13.051 values between 14.67 and 15.491 values between 15.49 and 16.301 values between 21.17 and 21.991 values between 22.80 and 23.61offnegativepositiveoften

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. lest
  2. ")!=
  3. ourselves
  4. "),"
  5. ")==
  6. лиш
  7. "]=="
  8. ");//
  9. ']]['
  10. /Desktop

Suppresses

  1. ,
  2. ,
  3. ,',
  4. ,C
  5. ,...⏎⏎
  6. ,F
  7. ,...⏎
  8. ،
  9. ,M
  10. ,S