Neuron Explorer

Qwen3-0.6B

comma linking two clauses after 'and'

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 a comma that comes right after a word like 'and' when it joins two full sentence parts, such as 'X, and, in some cases, Y'. When it was turned up, the model's writing fell into long loops of repeated words like 'and, and, and' or 'you, you, you'.

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

    ? There are only three species of venomous snakes living in the wilds of New York (many other kinds may be found in the homes of private individuals and,

  2. #2 peak

    briquettes, known as yeontan. For decades, these briquettes kept Korean houses warm in winter. They were seen as a valuable commodity and, at some point, an embodiment of technical progress.⏎Heating has always been a

  3. #3 peak

    greater freedom in operations, in return for higher performance goals and academic results specified in their contract, or “charter,” made and agreed upon by school leaders and, in IDEA’s case East Baton Rouge district officials.⏎How is IDEA funded and are

  4. #4 peak

    to land in Plymouth previously, employing four aircraft and dozens of naval ships to mark the way. What marked this flight out was that it was non-stop, and, taking less than 72 hours was eligible for the Daily Mail prize.⏎There

  5. #5 peak

    soccer with miniscule figurines in painted-on soccer outfits, mounted on rods that are manipulated from the sides of the table. You can move these rods and, thereby, the “players” right and left to the field and rotate them

  6. #6 peak

    .⏎- In the case of Col. Shaw, it seems that he chose to live his life a certain way (leading his African American troops into battle) and, now that he's dead, he's off on the sidelines, checking out all

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 value of $ \sum_{k=0}^{n^{2} r^{2} l^{2} t^{2} \ 0 1^{2} 1 0 1 2 0

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

    deal with every issue on the summit agenda. What follows are a few examples of the challenges and innovations under discussion.⏎Access to Education⏎At the most fundamental level, the world has still not achieved the commitment made by the world’s governments to provide

  2. #2 peak

    things that ordinary people just can’t!”⏎The appeal of such a dismissive explanation is that it excuses us from living, and perhaps even dying, if necessary, like Father Delp<|endoftext|>Hachiman is in the far south-east of Kh

  3. #3 peak

    ogy, which explores issues relevant to learning and teaching mathematics-in-context, thus providing a valuable practical tool for general mathematics education research. The most important goal, then, is to exemplify the merging of the theory with practice and the subsequent applications to

  4. #4 peak

    put together from Continental Europe, British Isles, and American building blocks in Southeastern Pennsylvania. While this culture is a unit linguistically, and in most other ways, it is divided down the middle by religion. The "two worlds" are those

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. L15:1615 excites
  2. L14:35 suppresses
  3. L15:241 excites
  4. L9:77 excites commas in citations and name lists
  5. L4:3 excites commas after citation fields
  6. L5:16 excites
  7. L10:856 excites
  8. L11:154 suppresses commas after citation numbers
  9. L6:250 excites
  10. L9:1113 suppresses
  11. L14:51 excites colon that introduces a statement
  12. L0:2921 suppresses
  13. L10:335 excites commas in catalog, date, and list metadata
  14. L7:288 excites
  15. L15:1213 suppresses
  16. L15:1772 suppresses
  17. L5:454 suppresses
  18. L10:1459 suppresses
  19. L3:71 suppresses
  20. L15:123 suppresses
  21. L15:1128 excites
  22. L2:66 excites
  23. L1:1 excites
  24. L15:630 suppresses

Feeds later layers

  1. L17:1772 suppresses
  2. L18:1394 suppresses
  3. L26:236 excites
  4. L17:1405 suppresses
  5. L19:682 excites
  6. L21:700 suppresses comma before a new clause
  7. L17:116 excites
  8. L17:757 excites
  9. L19:1036 excites
  10. L17:2001 suppresses
  11. L27:191 excites
  12. L17:348 excites
  13. L27:86 suppresses
  14. L19:726 suppresses comma inside large numbers
  15. L17:2413 suppresses
  16. L26:22 suppresses the 'res' fragment inside words
  17. L22:368 excites
  18. L20:173 excites
  19. L23:184 suppresses
  20. L25:328 excites
  21. L19:1506 suppresses
  22. L25:438 excites
  23. L19:2645 suppresses commas in the middle of a sentence
  24. L18:862 excites hedges like 'it turns out' and 'if any'

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 -6.82 and -6.271 values between -6.27 and -5.732 values between -5.73 and -5.183 values between -5.18 and -4.636 values between -4.63 and -4.0911 values between -4.09 and -3.5418 values between -3.54 and -3.0017 values between -3.00 and -2.4536 values between -2.45 and -1.9146 values between -1.91 and -1.3694 values between -1.36 and -0.82165 values between -0.82 and -0.271469 values between -0.27 and 0.281936 values between 0.28 and 0.8261 values between 0.82 and 1.3716 values between 1.37 and 1.919 values between 1.91 and 2.464 values between 2.46 and 3.004 values between 3.00 and 3.554 values between 3.55 and 4.101 values between 4.10 and 4.641 values between 4.64 and 5.191 values between 5.19 and 5.731 values between 6.28 and 6.821 values between 14.46 and 15.01offnegativepositiveoften

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. ,
  2. #,
  3. -down
  4. !,
  5. ,
  6. ees
  7. ,在
  8. 下去
  9. ,#
  10. geois

Suppresses

  1. можете
  2. '),('
  3. 냅
  4. 他们的
  5. 请您
  6. 除非
  7. 但是如果
  8. ⏎ ⏎ ⏎
  9. 您可以
  10. 您的